KVM: VMX: Intercept RDPMC
[linux-3.10.git] / arch / x86 / kvm / vmx.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "irq.h"
20 #include "mmu.h"
21 #include "cpuid.h"
22
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/ftrace_event.h>
31 #include <linux/slab.h>
32 #include <linux/tboot.h>
33 #include "kvm_cache_regs.h"
34 #include "x86.h"
35
36 #include <asm/io.h>
37 #include <asm/desc.h>
38 #include <asm/vmx.h>
39 #include <asm/virtext.h>
40 #include <asm/mce.h>
41 #include <asm/i387.h>
42 #include <asm/xcr.h>
43 #include <asm/perf_event.h>
44
45 #include "trace.h"
46
47 #define __ex(x) __kvm_handle_fault_on_reboot(x)
48 #define __ex_clear(x, reg) \
49         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
50
51 MODULE_AUTHOR("Qumranet");
52 MODULE_LICENSE("GPL");
53
54 static int __read_mostly enable_vpid = 1;
55 module_param_named(vpid, enable_vpid, bool, 0444);
56
57 static int __read_mostly flexpriority_enabled = 1;
58 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
59
60 static int __read_mostly enable_ept = 1;
61 module_param_named(ept, enable_ept, bool, S_IRUGO);
62
63 static int __read_mostly enable_unrestricted_guest = 1;
64 module_param_named(unrestricted_guest,
65                         enable_unrestricted_guest, bool, S_IRUGO);
66
67 static int __read_mostly emulate_invalid_guest_state = 0;
68 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
69
70 static int __read_mostly vmm_exclusive = 1;
71 module_param(vmm_exclusive, bool, S_IRUGO);
72
73 static int __read_mostly yield_on_hlt = 1;
74 module_param(yield_on_hlt, bool, S_IRUGO);
75
76 static int __read_mostly fasteoi = 1;
77 module_param(fasteoi, bool, S_IRUGO);
78
79 /*
80  * If nested=1, nested virtualization is supported, i.e., guests may use
81  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
82  * use VMX instructions.
83  */
84 static int __read_mostly nested = 0;
85 module_param(nested, bool, S_IRUGO);
86
87 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
88         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
89 #define KVM_GUEST_CR0_MASK                                              \
90         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
91 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
92         (X86_CR0_WP | X86_CR0_NE)
93 #define KVM_VM_CR0_ALWAYS_ON                                            \
94         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
95 #define KVM_CR4_GUEST_OWNED_BITS                                      \
96         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
97          | X86_CR4_OSXMMEXCPT)
98
99 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
100 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
101
102 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
103
104 /*
105  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
106  * ple_gap:    upper bound on the amount of time between two successive
107  *             executions of PAUSE in a loop. Also indicate if ple enabled.
108  *             According to test, this time is usually smaller than 128 cycles.
109  * ple_window: upper bound on the amount of time a guest is allowed to execute
110  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
111  *             less than 2^12 cycles
112  * Time is measured based on a counter that runs at the same rate as the TSC,
113  * refer SDM volume 3b section 21.6.13 & 22.1.3.
114  */
115 #define KVM_VMX_DEFAULT_PLE_GAP    128
116 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
117 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
118 module_param(ple_gap, int, S_IRUGO);
119
120 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
121 module_param(ple_window, int, S_IRUGO);
122
123 #define NR_AUTOLOAD_MSRS 8
124 #define VMCS02_POOL_SIZE 1
125
126 struct vmcs {
127         u32 revision_id;
128         u32 abort;
129         char data[0];
130 };
131
132 /*
133  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
134  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
135  * loaded on this CPU (so we can clear them if the CPU goes down).
136  */
137 struct loaded_vmcs {
138         struct vmcs *vmcs;
139         int cpu;
140         int launched;
141         struct list_head loaded_vmcss_on_cpu_link;
142 };
143
144 struct shared_msr_entry {
145         unsigned index;
146         u64 data;
147         u64 mask;
148 };
149
150 /*
151  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
152  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
153  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
154  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
155  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
156  * More than one of these structures may exist, if L1 runs multiple L2 guests.
157  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
158  * underlying hardware which will be used to run L2.
159  * This structure is packed to ensure that its layout is identical across
160  * machines (necessary for live migration).
161  * If there are changes in this struct, VMCS12_REVISION must be changed.
162  */
163 typedef u64 natural_width;
164 struct __packed vmcs12 {
165         /* According to the Intel spec, a VMCS region must start with the
166          * following two fields. Then follow implementation-specific data.
167          */
168         u32 revision_id;
169         u32 abort;
170
171         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
172         u32 padding[7]; /* room for future expansion */
173
174         u64 io_bitmap_a;
175         u64 io_bitmap_b;
176         u64 msr_bitmap;
177         u64 vm_exit_msr_store_addr;
178         u64 vm_exit_msr_load_addr;
179         u64 vm_entry_msr_load_addr;
180         u64 tsc_offset;
181         u64 virtual_apic_page_addr;
182         u64 apic_access_addr;
183         u64 ept_pointer;
184         u64 guest_physical_address;
185         u64 vmcs_link_pointer;
186         u64 guest_ia32_debugctl;
187         u64 guest_ia32_pat;
188         u64 guest_ia32_efer;
189         u64 guest_ia32_perf_global_ctrl;
190         u64 guest_pdptr0;
191         u64 guest_pdptr1;
192         u64 guest_pdptr2;
193         u64 guest_pdptr3;
194         u64 host_ia32_pat;
195         u64 host_ia32_efer;
196         u64 host_ia32_perf_global_ctrl;
197         u64 padding64[8]; /* room for future expansion */
198         /*
199          * To allow migration of L1 (complete with its L2 guests) between
200          * machines of different natural widths (32 or 64 bit), we cannot have
201          * unsigned long fields with no explict size. We use u64 (aliased
202          * natural_width) instead. Luckily, x86 is little-endian.
203          */
204         natural_width cr0_guest_host_mask;
205         natural_width cr4_guest_host_mask;
206         natural_width cr0_read_shadow;
207         natural_width cr4_read_shadow;
208         natural_width cr3_target_value0;
209         natural_width cr3_target_value1;
210         natural_width cr3_target_value2;
211         natural_width cr3_target_value3;
212         natural_width exit_qualification;
213         natural_width guest_linear_address;
214         natural_width guest_cr0;
215         natural_width guest_cr3;
216         natural_width guest_cr4;
217         natural_width guest_es_base;
218         natural_width guest_cs_base;
219         natural_width guest_ss_base;
220         natural_width guest_ds_base;
221         natural_width guest_fs_base;
222         natural_width guest_gs_base;
223         natural_width guest_ldtr_base;
224         natural_width guest_tr_base;
225         natural_width guest_gdtr_base;
226         natural_width guest_idtr_base;
227         natural_width guest_dr7;
228         natural_width guest_rsp;
229         natural_width guest_rip;
230         natural_width guest_rflags;
231         natural_width guest_pending_dbg_exceptions;
232         natural_width guest_sysenter_esp;
233         natural_width guest_sysenter_eip;
234         natural_width host_cr0;
235         natural_width host_cr3;
236         natural_width host_cr4;
237         natural_width host_fs_base;
238         natural_width host_gs_base;
239         natural_width host_tr_base;
240         natural_width host_gdtr_base;
241         natural_width host_idtr_base;
242         natural_width host_ia32_sysenter_esp;
243         natural_width host_ia32_sysenter_eip;
244         natural_width host_rsp;
245         natural_width host_rip;
246         natural_width paddingl[8]; /* room for future expansion */
247         u32 pin_based_vm_exec_control;
248         u32 cpu_based_vm_exec_control;
249         u32 exception_bitmap;
250         u32 page_fault_error_code_mask;
251         u32 page_fault_error_code_match;
252         u32 cr3_target_count;
253         u32 vm_exit_controls;
254         u32 vm_exit_msr_store_count;
255         u32 vm_exit_msr_load_count;
256         u32 vm_entry_controls;
257         u32 vm_entry_msr_load_count;
258         u32 vm_entry_intr_info_field;
259         u32 vm_entry_exception_error_code;
260         u32 vm_entry_instruction_len;
261         u32 tpr_threshold;
262         u32 secondary_vm_exec_control;
263         u32 vm_instruction_error;
264         u32 vm_exit_reason;
265         u32 vm_exit_intr_info;
266         u32 vm_exit_intr_error_code;
267         u32 idt_vectoring_info_field;
268         u32 idt_vectoring_error_code;
269         u32 vm_exit_instruction_len;
270         u32 vmx_instruction_info;
271         u32 guest_es_limit;
272         u32 guest_cs_limit;
273         u32 guest_ss_limit;
274         u32 guest_ds_limit;
275         u32 guest_fs_limit;
276         u32 guest_gs_limit;
277         u32 guest_ldtr_limit;
278         u32 guest_tr_limit;
279         u32 guest_gdtr_limit;
280         u32 guest_idtr_limit;
281         u32 guest_es_ar_bytes;
282         u32 guest_cs_ar_bytes;
283         u32 guest_ss_ar_bytes;
284         u32 guest_ds_ar_bytes;
285         u32 guest_fs_ar_bytes;
286         u32 guest_gs_ar_bytes;
287         u32 guest_ldtr_ar_bytes;
288         u32 guest_tr_ar_bytes;
289         u32 guest_interruptibility_info;
290         u32 guest_activity_state;
291         u32 guest_sysenter_cs;
292         u32 host_ia32_sysenter_cs;
293         u32 padding32[8]; /* room for future expansion */
294         u16 virtual_processor_id;
295         u16 guest_es_selector;
296         u16 guest_cs_selector;
297         u16 guest_ss_selector;
298         u16 guest_ds_selector;
299         u16 guest_fs_selector;
300         u16 guest_gs_selector;
301         u16 guest_ldtr_selector;
302         u16 guest_tr_selector;
303         u16 host_es_selector;
304         u16 host_cs_selector;
305         u16 host_ss_selector;
306         u16 host_ds_selector;
307         u16 host_fs_selector;
308         u16 host_gs_selector;
309         u16 host_tr_selector;
310 };
311
312 /*
313  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
314  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
315  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
316  */
317 #define VMCS12_REVISION 0x11e57ed0
318
319 /*
320  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
321  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
322  * current implementation, 4K are reserved to avoid future complications.
323  */
324 #define VMCS12_SIZE 0x1000
325
326 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
327 struct vmcs02_list {
328         struct list_head list;
329         gpa_t vmptr;
330         struct loaded_vmcs vmcs02;
331 };
332
333 /*
334  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
335  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
336  */
337 struct nested_vmx {
338         /* Has the level1 guest done vmxon? */
339         bool vmxon;
340
341         /* The guest-physical address of the current VMCS L1 keeps for L2 */
342         gpa_t current_vmptr;
343         /* The host-usable pointer to the above */
344         struct page *current_vmcs12_page;
345         struct vmcs12 *current_vmcs12;
346
347         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
348         struct list_head vmcs02_pool;
349         int vmcs02_num;
350         u64 vmcs01_tsc_offset;
351         /* L2 must run next, and mustn't decide to exit to L1. */
352         bool nested_run_pending;
353         /*
354          * Guest pages referred to in vmcs02 with host-physical pointers, so
355          * we must keep them pinned while L2 runs.
356          */
357         struct page *apic_access_page;
358 };
359
360 struct vcpu_vmx {
361         struct kvm_vcpu       vcpu;
362         unsigned long         host_rsp;
363         u8                    fail;
364         u8                    cpl;
365         bool                  nmi_known_unmasked;
366         u32                   exit_intr_info;
367         u32                   idt_vectoring_info;
368         ulong                 rflags;
369         struct shared_msr_entry *guest_msrs;
370         int                   nmsrs;
371         int                   save_nmsrs;
372 #ifdef CONFIG_X86_64
373         u64                   msr_host_kernel_gs_base;
374         u64                   msr_guest_kernel_gs_base;
375 #endif
376         /*
377          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
378          * non-nested (L1) guest, it always points to vmcs01. For a nested
379          * guest (L2), it points to a different VMCS.
380          */
381         struct loaded_vmcs    vmcs01;
382         struct loaded_vmcs   *loaded_vmcs;
383         bool                  __launched; /* temporary, used in vmx_vcpu_run */
384         struct msr_autoload {
385                 unsigned nr;
386                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
387                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
388         } msr_autoload;
389         struct {
390                 int           loaded;
391                 u16           fs_sel, gs_sel, ldt_sel;
392                 int           gs_ldt_reload_needed;
393                 int           fs_reload_needed;
394         } host_state;
395         struct {
396                 int vm86_active;
397                 ulong save_rflags;
398                 struct kvm_save_segment {
399                         u16 selector;
400                         unsigned long base;
401                         u32 limit;
402                         u32 ar;
403                 } tr, es, ds, fs, gs;
404         } rmode;
405         struct {
406                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
407                 struct kvm_save_segment seg[8];
408         } segment_cache;
409         int vpid;
410         bool emulation_required;
411
412         /* Support for vnmi-less CPUs */
413         int soft_vnmi_blocked;
414         ktime_t entry_time;
415         s64 vnmi_blocked_time;
416         u32 exit_reason;
417
418         bool rdtscp_enabled;
419
420         /* Support for a guest hypervisor (nested VMX) */
421         struct nested_vmx nested;
422 };
423
424 enum segment_cache_field {
425         SEG_FIELD_SEL = 0,
426         SEG_FIELD_BASE = 1,
427         SEG_FIELD_LIMIT = 2,
428         SEG_FIELD_AR = 3,
429
430         SEG_FIELD_NR = 4
431 };
432
433 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
434 {
435         return container_of(vcpu, struct vcpu_vmx, vcpu);
436 }
437
438 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
439 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
440 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
441                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
442
443 static unsigned short vmcs_field_to_offset_table[] = {
444         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
445         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
446         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
447         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
448         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
449         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
450         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
451         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
452         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
453         FIELD(HOST_ES_SELECTOR, host_es_selector),
454         FIELD(HOST_CS_SELECTOR, host_cs_selector),
455         FIELD(HOST_SS_SELECTOR, host_ss_selector),
456         FIELD(HOST_DS_SELECTOR, host_ds_selector),
457         FIELD(HOST_FS_SELECTOR, host_fs_selector),
458         FIELD(HOST_GS_SELECTOR, host_gs_selector),
459         FIELD(HOST_TR_SELECTOR, host_tr_selector),
460         FIELD64(IO_BITMAP_A, io_bitmap_a),
461         FIELD64(IO_BITMAP_B, io_bitmap_b),
462         FIELD64(MSR_BITMAP, msr_bitmap),
463         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
464         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
465         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
466         FIELD64(TSC_OFFSET, tsc_offset),
467         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
468         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
469         FIELD64(EPT_POINTER, ept_pointer),
470         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
471         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
472         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
473         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
474         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
475         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
476         FIELD64(GUEST_PDPTR0, guest_pdptr0),
477         FIELD64(GUEST_PDPTR1, guest_pdptr1),
478         FIELD64(GUEST_PDPTR2, guest_pdptr2),
479         FIELD64(GUEST_PDPTR3, guest_pdptr3),
480         FIELD64(HOST_IA32_PAT, host_ia32_pat),
481         FIELD64(HOST_IA32_EFER, host_ia32_efer),
482         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
483         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
484         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
485         FIELD(EXCEPTION_BITMAP, exception_bitmap),
486         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
487         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
488         FIELD(CR3_TARGET_COUNT, cr3_target_count),
489         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
490         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
491         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
492         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
493         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
494         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
495         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
496         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
497         FIELD(TPR_THRESHOLD, tpr_threshold),
498         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
499         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
500         FIELD(VM_EXIT_REASON, vm_exit_reason),
501         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
502         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
503         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
504         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
505         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
506         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
507         FIELD(GUEST_ES_LIMIT, guest_es_limit),
508         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
509         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
510         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
511         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
512         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
513         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
514         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
515         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
516         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
517         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
518         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
519         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
520         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
521         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
522         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
523         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
524         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
525         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
526         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
527         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
528         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
529         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
530         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
531         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
532         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
533         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
534         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
535         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
536         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
537         FIELD(EXIT_QUALIFICATION, exit_qualification),
538         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
539         FIELD(GUEST_CR0, guest_cr0),
540         FIELD(GUEST_CR3, guest_cr3),
541         FIELD(GUEST_CR4, guest_cr4),
542         FIELD(GUEST_ES_BASE, guest_es_base),
543         FIELD(GUEST_CS_BASE, guest_cs_base),
544         FIELD(GUEST_SS_BASE, guest_ss_base),
545         FIELD(GUEST_DS_BASE, guest_ds_base),
546         FIELD(GUEST_FS_BASE, guest_fs_base),
547         FIELD(GUEST_GS_BASE, guest_gs_base),
548         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
549         FIELD(GUEST_TR_BASE, guest_tr_base),
550         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
551         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
552         FIELD(GUEST_DR7, guest_dr7),
553         FIELD(GUEST_RSP, guest_rsp),
554         FIELD(GUEST_RIP, guest_rip),
555         FIELD(GUEST_RFLAGS, guest_rflags),
556         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
557         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
558         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
559         FIELD(HOST_CR0, host_cr0),
560         FIELD(HOST_CR3, host_cr3),
561         FIELD(HOST_CR4, host_cr4),
562         FIELD(HOST_FS_BASE, host_fs_base),
563         FIELD(HOST_GS_BASE, host_gs_base),
564         FIELD(HOST_TR_BASE, host_tr_base),
565         FIELD(HOST_GDTR_BASE, host_gdtr_base),
566         FIELD(HOST_IDTR_BASE, host_idtr_base),
567         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
568         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
569         FIELD(HOST_RSP, host_rsp),
570         FIELD(HOST_RIP, host_rip),
571 };
572 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
573
574 static inline short vmcs_field_to_offset(unsigned long field)
575 {
576         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
577                 return -1;
578         return vmcs_field_to_offset_table[field];
579 }
580
581 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
582 {
583         return to_vmx(vcpu)->nested.current_vmcs12;
584 }
585
586 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
587 {
588         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
589         if (is_error_page(page)) {
590                 kvm_release_page_clean(page);
591                 return NULL;
592         }
593         return page;
594 }
595
596 static void nested_release_page(struct page *page)
597 {
598         kvm_release_page_dirty(page);
599 }
600
601 static void nested_release_page_clean(struct page *page)
602 {
603         kvm_release_page_clean(page);
604 }
605
606 static u64 construct_eptp(unsigned long root_hpa);
607 static void kvm_cpu_vmxon(u64 addr);
608 static void kvm_cpu_vmxoff(void);
609 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
610 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
611
612 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
613 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
614 /*
615  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
616  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
617  */
618 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
619 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
620
621 static unsigned long *vmx_io_bitmap_a;
622 static unsigned long *vmx_io_bitmap_b;
623 static unsigned long *vmx_msr_bitmap_legacy;
624 static unsigned long *vmx_msr_bitmap_longmode;
625
626 static bool cpu_has_load_ia32_efer;
627 static bool cpu_has_load_perf_global_ctrl;
628
629 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
630 static DEFINE_SPINLOCK(vmx_vpid_lock);
631
632 static struct vmcs_config {
633         int size;
634         int order;
635         u32 revision_id;
636         u32 pin_based_exec_ctrl;
637         u32 cpu_based_exec_ctrl;
638         u32 cpu_based_2nd_exec_ctrl;
639         u32 vmexit_ctrl;
640         u32 vmentry_ctrl;
641 } vmcs_config;
642
643 static struct vmx_capability {
644         u32 ept;
645         u32 vpid;
646 } vmx_capability;
647
648 #define VMX_SEGMENT_FIELD(seg)                                  \
649         [VCPU_SREG_##seg] = {                                   \
650                 .selector = GUEST_##seg##_SELECTOR,             \
651                 .base = GUEST_##seg##_BASE,                     \
652                 .limit = GUEST_##seg##_LIMIT,                   \
653                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
654         }
655
656 static struct kvm_vmx_segment_field {
657         unsigned selector;
658         unsigned base;
659         unsigned limit;
660         unsigned ar_bytes;
661 } kvm_vmx_segment_fields[] = {
662         VMX_SEGMENT_FIELD(CS),
663         VMX_SEGMENT_FIELD(DS),
664         VMX_SEGMENT_FIELD(ES),
665         VMX_SEGMENT_FIELD(FS),
666         VMX_SEGMENT_FIELD(GS),
667         VMX_SEGMENT_FIELD(SS),
668         VMX_SEGMENT_FIELD(TR),
669         VMX_SEGMENT_FIELD(LDTR),
670 };
671
672 static u64 host_efer;
673
674 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
675
676 /*
677  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
678  * away by decrementing the array size.
679  */
680 static const u32 vmx_msr_index[] = {
681 #ifdef CONFIG_X86_64
682         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
683 #endif
684         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
685 };
686 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
687
688 static inline bool is_page_fault(u32 intr_info)
689 {
690         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
691                              INTR_INFO_VALID_MASK)) ==
692                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
693 }
694
695 static inline bool is_no_device(u32 intr_info)
696 {
697         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
698                              INTR_INFO_VALID_MASK)) ==
699                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
700 }
701
702 static inline bool is_invalid_opcode(u32 intr_info)
703 {
704         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
705                              INTR_INFO_VALID_MASK)) ==
706                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
707 }
708
709 static inline bool is_external_interrupt(u32 intr_info)
710 {
711         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
712                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
713 }
714
715 static inline bool is_machine_check(u32 intr_info)
716 {
717         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
718                              INTR_INFO_VALID_MASK)) ==
719                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
720 }
721
722 static inline bool cpu_has_vmx_msr_bitmap(void)
723 {
724         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
725 }
726
727 static inline bool cpu_has_vmx_tpr_shadow(void)
728 {
729         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
730 }
731
732 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
733 {
734         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
735 }
736
737 static inline bool cpu_has_secondary_exec_ctrls(void)
738 {
739         return vmcs_config.cpu_based_exec_ctrl &
740                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
741 }
742
743 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
744 {
745         return vmcs_config.cpu_based_2nd_exec_ctrl &
746                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
747 }
748
749 static inline bool cpu_has_vmx_flexpriority(void)
750 {
751         return cpu_has_vmx_tpr_shadow() &&
752                 cpu_has_vmx_virtualize_apic_accesses();
753 }
754
755 static inline bool cpu_has_vmx_ept_execute_only(void)
756 {
757         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
758 }
759
760 static inline bool cpu_has_vmx_eptp_uncacheable(void)
761 {
762         return vmx_capability.ept & VMX_EPTP_UC_BIT;
763 }
764
765 static inline bool cpu_has_vmx_eptp_writeback(void)
766 {
767         return vmx_capability.ept & VMX_EPTP_WB_BIT;
768 }
769
770 static inline bool cpu_has_vmx_ept_2m_page(void)
771 {
772         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
773 }
774
775 static inline bool cpu_has_vmx_ept_1g_page(void)
776 {
777         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
778 }
779
780 static inline bool cpu_has_vmx_ept_4levels(void)
781 {
782         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
783 }
784
785 static inline bool cpu_has_vmx_invept_individual_addr(void)
786 {
787         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
788 }
789
790 static inline bool cpu_has_vmx_invept_context(void)
791 {
792         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
793 }
794
795 static inline bool cpu_has_vmx_invept_global(void)
796 {
797         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
798 }
799
800 static inline bool cpu_has_vmx_invvpid_single(void)
801 {
802         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
803 }
804
805 static inline bool cpu_has_vmx_invvpid_global(void)
806 {
807         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
808 }
809
810 static inline bool cpu_has_vmx_ept(void)
811 {
812         return vmcs_config.cpu_based_2nd_exec_ctrl &
813                 SECONDARY_EXEC_ENABLE_EPT;
814 }
815
816 static inline bool cpu_has_vmx_unrestricted_guest(void)
817 {
818         return vmcs_config.cpu_based_2nd_exec_ctrl &
819                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
820 }
821
822 static inline bool cpu_has_vmx_ple(void)
823 {
824         return vmcs_config.cpu_based_2nd_exec_ctrl &
825                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
826 }
827
828 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
829 {
830         return flexpriority_enabled && irqchip_in_kernel(kvm);
831 }
832
833 static inline bool cpu_has_vmx_vpid(void)
834 {
835         return vmcs_config.cpu_based_2nd_exec_ctrl &
836                 SECONDARY_EXEC_ENABLE_VPID;
837 }
838
839 static inline bool cpu_has_vmx_rdtscp(void)
840 {
841         return vmcs_config.cpu_based_2nd_exec_ctrl &
842                 SECONDARY_EXEC_RDTSCP;
843 }
844
845 static inline bool cpu_has_virtual_nmis(void)
846 {
847         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
848 }
849
850 static inline bool cpu_has_vmx_wbinvd_exit(void)
851 {
852         return vmcs_config.cpu_based_2nd_exec_ctrl &
853                 SECONDARY_EXEC_WBINVD_EXITING;
854 }
855
856 static inline bool report_flexpriority(void)
857 {
858         return flexpriority_enabled;
859 }
860
861 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
862 {
863         return vmcs12->cpu_based_vm_exec_control & bit;
864 }
865
866 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
867 {
868         return (vmcs12->cpu_based_vm_exec_control &
869                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
870                 (vmcs12->secondary_vm_exec_control & bit);
871 }
872
873 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
874         struct kvm_vcpu *vcpu)
875 {
876         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
877 }
878
879 static inline bool is_exception(u32 intr_info)
880 {
881         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
882                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
883 }
884
885 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
886 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
887                         struct vmcs12 *vmcs12,
888                         u32 reason, unsigned long qualification);
889
890 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
891 {
892         int i;
893
894         for (i = 0; i < vmx->nmsrs; ++i)
895                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
896                         return i;
897         return -1;
898 }
899
900 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
901 {
902     struct {
903         u64 vpid : 16;
904         u64 rsvd : 48;
905         u64 gva;
906     } operand = { vpid, 0, gva };
907
908     asm volatile (__ex(ASM_VMX_INVVPID)
909                   /* CF==1 or ZF==1 --> rc = -1 */
910                   "; ja 1f ; ud2 ; 1:"
911                   : : "a"(&operand), "c"(ext) : "cc", "memory");
912 }
913
914 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
915 {
916         struct {
917                 u64 eptp, gpa;
918         } operand = {eptp, gpa};
919
920         asm volatile (__ex(ASM_VMX_INVEPT)
921                         /* CF==1 or ZF==1 --> rc = -1 */
922                         "; ja 1f ; ud2 ; 1:\n"
923                         : : "a" (&operand), "c" (ext) : "cc", "memory");
924 }
925
926 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
927 {
928         int i;
929
930         i = __find_msr_index(vmx, msr);
931         if (i >= 0)
932                 return &vmx->guest_msrs[i];
933         return NULL;
934 }
935
936 static void vmcs_clear(struct vmcs *vmcs)
937 {
938         u64 phys_addr = __pa(vmcs);
939         u8 error;
940
941         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
942                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
943                       : "cc", "memory");
944         if (error)
945                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
946                        vmcs, phys_addr);
947 }
948
949 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
950 {
951         vmcs_clear(loaded_vmcs->vmcs);
952         loaded_vmcs->cpu = -1;
953         loaded_vmcs->launched = 0;
954 }
955
956 static void vmcs_load(struct vmcs *vmcs)
957 {
958         u64 phys_addr = __pa(vmcs);
959         u8 error;
960
961         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
962                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
963                         : "cc", "memory");
964         if (error)
965                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
966                        vmcs, phys_addr);
967 }
968
969 static void __loaded_vmcs_clear(void *arg)
970 {
971         struct loaded_vmcs *loaded_vmcs = arg;
972         int cpu = raw_smp_processor_id();
973
974         if (loaded_vmcs->cpu != cpu)
975                 return; /* vcpu migration can race with cpu offline */
976         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
977                 per_cpu(current_vmcs, cpu) = NULL;
978         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
979         loaded_vmcs_init(loaded_vmcs);
980 }
981
982 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
983 {
984         if (loaded_vmcs->cpu != -1)
985                 smp_call_function_single(
986                         loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
987 }
988
989 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
990 {
991         if (vmx->vpid == 0)
992                 return;
993
994         if (cpu_has_vmx_invvpid_single())
995                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
996 }
997
998 static inline void vpid_sync_vcpu_global(void)
999 {
1000         if (cpu_has_vmx_invvpid_global())
1001                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1002 }
1003
1004 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1005 {
1006         if (cpu_has_vmx_invvpid_single())
1007                 vpid_sync_vcpu_single(vmx);
1008         else
1009                 vpid_sync_vcpu_global();
1010 }
1011
1012 static inline void ept_sync_global(void)
1013 {
1014         if (cpu_has_vmx_invept_global())
1015                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1016 }
1017
1018 static inline void ept_sync_context(u64 eptp)
1019 {
1020         if (enable_ept) {
1021                 if (cpu_has_vmx_invept_context())
1022                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1023                 else
1024                         ept_sync_global();
1025         }
1026 }
1027
1028 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1029 {
1030         if (enable_ept) {
1031                 if (cpu_has_vmx_invept_individual_addr())
1032                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1033                                         eptp, gpa);
1034                 else
1035                         ept_sync_context(eptp);
1036         }
1037 }
1038
1039 static __always_inline unsigned long vmcs_readl(unsigned long field)
1040 {
1041         unsigned long value;
1042
1043         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1044                       : "=a"(value) : "d"(field) : "cc");
1045         return value;
1046 }
1047
1048 static __always_inline u16 vmcs_read16(unsigned long field)
1049 {
1050         return vmcs_readl(field);
1051 }
1052
1053 static __always_inline u32 vmcs_read32(unsigned long field)
1054 {
1055         return vmcs_readl(field);
1056 }
1057
1058 static __always_inline u64 vmcs_read64(unsigned long field)
1059 {
1060 #ifdef CONFIG_X86_64
1061         return vmcs_readl(field);
1062 #else
1063         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1064 #endif
1065 }
1066
1067 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1068 {
1069         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1070                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1071         dump_stack();
1072 }
1073
1074 static void vmcs_writel(unsigned long field, unsigned long value)
1075 {
1076         u8 error;
1077
1078         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1079                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1080         if (unlikely(error))
1081                 vmwrite_error(field, value);
1082 }
1083
1084 static void vmcs_write16(unsigned long field, u16 value)
1085 {
1086         vmcs_writel(field, value);
1087 }
1088
1089 static void vmcs_write32(unsigned long field, u32 value)
1090 {
1091         vmcs_writel(field, value);
1092 }
1093
1094 static void vmcs_write64(unsigned long field, u64 value)
1095 {
1096         vmcs_writel(field, value);
1097 #ifndef CONFIG_X86_64
1098         asm volatile ("");
1099         vmcs_writel(field+1, value >> 32);
1100 #endif
1101 }
1102
1103 static void vmcs_clear_bits(unsigned long field, u32 mask)
1104 {
1105         vmcs_writel(field, vmcs_readl(field) & ~mask);
1106 }
1107
1108 static void vmcs_set_bits(unsigned long field, u32 mask)
1109 {
1110         vmcs_writel(field, vmcs_readl(field) | mask);
1111 }
1112
1113 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1114 {
1115         vmx->segment_cache.bitmask = 0;
1116 }
1117
1118 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1119                                        unsigned field)
1120 {
1121         bool ret;
1122         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1123
1124         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1125                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1126                 vmx->segment_cache.bitmask = 0;
1127         }
1128         ret = vmx->segment_cache.bitmask & mask;
1129         vmx->segment_cache.bitmask |= mask;
1130         return ret;
1131 }
1132
1133 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1134 {
1135         u16 *p = &vmx->segment_cache.seg[seg].selector;
1136
1137         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1138                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1139         return *p;
1140 }
1141
1142 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1143 {
1144         ulong *p = &vmx->segment_cache.seg[seg].base;
1145
1146         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1147                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1148         return *p;
1149 }
1150
1151 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1152 {
1153         u32 *p = &vmx->segment_cache.seg[seg].limit;
1154
1155         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1156                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1157         return *p;
1158 }
1159
1160 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1161 {
1162         u32 *p = &vmx->segment_cache.seg[seg].ar;
1163
1164         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1165                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1166         return *p;
1167 }
1168
1169 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1170 {
1171         u32 eb;
1172
1173         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1174              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1175         if ((vcpu->guest_debug &
1176              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1177             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1178                 eb |= 1u << BP_VECTOR;
1179         if (to_vmx(vcpu)->rmode.vm86_active)
1180                 eb = ~0;
1181         if (enable_ept)
1182                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1183         if (vcpu->fpu_active)
1184                 eb &= ~(1u << NM_VECTOR);
1185
1186         /* When we are running a nested L2 guest and L1 specified for it a
1187          * certain exception bitmap, we must trap the same exceptions and pass
1188          * them to L1. When running L2, we will only handle the exceptions
1189          * specified above if L1 did not want them.
1190          */
1191         if (is_guest_mode(vcpu))
1192                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1193
1194         vmcs_write32(EXCEPTION_BITMAP, eb);
1195 }
1196
1197 static void clear_atomic_switch_msr_special(unsigned long entry,
1198                 unsigned long exit)
1199 {
1200         vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1201         vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1202 }
1203
1204 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1205 {
1206         unsigned i;
1207         struct msr_autoload *m = &vmx->msr_autoload;
1208
1209         switch (msr) {
1210         case MSR_EFER:
1211                 if (cpu_has_load_ia32_efer) {
1212                         clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1213                                         VM_EXIT_LOAD_IA32_EFER);
1214                         return;
1215                 }
1216                 break;
1217         case MSR_CORE_PERF_GLOBAL_CTRL:
1218                 if (cpu_has_load_perf_global_ctrl) {
1219                         clear_atomic_switch_msr_special(
1220                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1221                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1222                         return;
1223                 }
1224                 break;
1225         }
1226
1227         for (i = 0; i < m->nr; ++i)
1228                 if (m->guest[i].index == msr)
1229                         break;
1230
1231         if (i == m->nr)
1232                 return;
1233         --m->nr;
1234         m->guest[i] = m->guest[m->nr];
1235         m->host[i] = m->host[m->nr];
1236         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1237         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1238 }
1239
1240 static void add_atomic_switch_msr_special(unsigned long entry,
1241                 unsigned long exit, unsigned long guest_val_vmcs,
1242                 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1243 {
1244         vmcs_write64(guest_val_vmcs, guest_val);
1245         vmcs_write64(host_val_vmcs, host_val);
1246         vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1247         vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1248 }
1249
1250 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1251                                   u64 guest_val, u64 host_val)
1252 {
1253         unsigned i;
1254         struct msr_autoload *m = &vmx->msr_autoload;
1255
1256         switch (msr) {
1257         case MSR_EFER:
1258                 if (cpu_has_load_ia32_efer) {
1259                         add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1260                                         VM_EXIT_LOAD_IA32_EFER,
1261                                         GUEST_IA32_EFER,
1262                                         HOST_IA32_EFER,
1263                                         guest_val, host_val);
1264                         return;
1265                 }
1266                 break;
1267         case MSR_CORE_PERF_GLOBAL_CTRL:
1268                 if (cpu_has_load_perf_global_ctrl) {
1269                         add_atomic_switch_msr_special(
1270                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1271                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1272                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1273                                         HOST_IA32_PERF_GLOBAL_CTRL,
1274                                         guest_val, host_val);
1275                         return;
1276                 }
1277                 break;
1278         }
1279
1280         for (i = 0; i < m->nr; ++i)
1281                 if (m->guest[i].index == msr)
1282                         break;
1283
1284         if (i == NR_AUTOLOAD_MSRS) {
1285                 printk_once(KERN_WARNING"Not enough mst switch entries. "
1286                                 "Can't add msr %x\n", msr);
1287                 return;
1288         } else if (i == m->nr) {
1289                 ++m->nr;
1290                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1291                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1292         }
1293
1294         m->guest[i].index = msr;
1295         m->guest[i].value = guest_val;
1296         m->host[i].index = msr;
1297         m->host[i].value = host_val;
1298 }
1299
1300 static void reload_tss(void)
1301 {
1302         /*
1303          * VT restores TR but not its size.  Useless.
1304          */
1305         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1306         struct desc_struct *descs;
1307
1308         descs = (void *)gdt->address;
1309         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1310         load_TR_desc();
1311 }
1312
1313 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1314 {
1315         u64 guest_efer;
1316         u64 ignore_bits;
1317
1318         guest_efer = vmx->vcpu.arch.efer;
1319
1320         /*
1321          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1322          * outside long mode
1323          */
1324         ignore_bits = EFER_NX | EFER_SCE;
1325 #ifdef CONFIG_X86_64
1326         ignore_bits |= EFER_LMA | EFER_LME;
1327         /* SCE is meaningful only in long mode on Intel */
1328         if (guest_efer & EFER_LMA)
1329                 ignore_bits &= ~(u64)EFER_SCE;
1330 #endif
1331         guest_efer &= ~ignore_bits;
1332         guest_efer |= host_efer & ignore_bits;
1333         vmx->guest_msrs[efer_offset].data = guest_efer;
1334         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1335
1336         clear_atomic_switch_msr(vmx, MSR_EFER);
1337         /* On ept, can't emulate nx, and must switch nx atomically */
1338         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1339                 guest_efer = vmx->vcpu.arch.efer;
1340                 if (!(guest_efer & EFER_LMA))
1341                         guest_efer &= ~EFER_LME;
1342                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1343                 return false;
1344         }
1345
1346         return true;
1347 }
1348
1349 static unsigned long segment_base(u16 selector)
1350 {
1351         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1352         struct desc_struct *d;
1353         unsigned long table_base;
1354         unsigned long v;
1355
1356         if (!(selector & ~3))
1357                 return 0;
1358
1359         table_base = gdt->address;
1360
1361         if (selector & 4) {           /* from ldt */
1362                 u16 ldt_selector = kvm_read_ldt();
1363
1364                 if (!(ldt_selector & ~3))
1365                         return 0;
1366
1367                 table_base = segment_base(ldt_selector);
1368         }
1369         d = (struct desc_struct *)(table_base + (selector & ~7));
1370         v = get_desc_base(d);
1371 #ifdef CONFIG_X86_64
1372        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1373                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1374 #endif
1375         return v;
1376 }
1377
1378 static inline unsigned long kvm_read_tr_base(void)
1379 {
1380         u16 tr;
1381         asm("str %0" : "=g"(tr));
1382         return segment_base(tr);
1383 }
1384
1385 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1386 {
1387         struct vcpu_vmx *vmx = to_vmx(vcpu);
1388         int i;
1389
1390         if (vmx->host_state.loaded)
1391                 return;
1392
1393         vmx->host_state.loaded = 1;
1394         /*
1395          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1396          * allow segment selectors with cpl > 0 or ti == 1.
1397          */
1398         vmx->host_state.ldt_sel = kvm_read_ldt();
1399         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1400         savesegment(fs, vmx->host_state.fs_sel);
1401         if (!(vmx->host_state.fs_sel & 7)) {
1402                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1403                 vmx->host_state.fs_reload_needed = 0;
1404         } else {
1405                 vmcs_write16(HOST_FS_SELECTOR, 0);
1406                 vmx->host_state.fs_reload_needed = 1;
1407         }
1408         savesegment(gs, vmx->host_state.gs_sel);
1409         if (!(vmx->host_state.gs_sel & 7))
1410                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1411         else {
1412                 vmcs_write16(HOST_GS_SELECTOR, 0);
1413                 vmx->host_state.gs_ldt_reload_needed = 1;
1414         }
1415
1416 #ifdef CONFIG_X86_64
1417         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1418         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1419 #else
1420         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1421         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1422 #endif
1423
1424 #ifdef CONFIG_X86_64
1425         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1426         if (is_long_mode(&vmx->vcpu))
1427                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1428 #endif
1429         for (i = 0; i < vmx->save_nmsrs; ++i)
1430                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1431                                    vmx->guest_msrs[i].data,
1432                                    vmx->guest_msrs[i].mask);
1433 }
1434
1435 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1436 {
1437         if (!vmx->host_state.loaded)
1438                 return;
1439
1440         ++vmx->vcpu.stat.host_state_reload;
1441         vmx->host_state.loaded = 0;
1442 #ifdef CONFIG_X86_64
1443         if (is_long_mode(&vmx->vcpu))
1444                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1445 #endif
1446         if (vmx->host_state.gs_ldt_reload_needed) {
1447                 kvm_load_ldt(vmx->host_state.ldt_sel);
1448 #ifdef CONFIG_X86_64
1449                 load_gs_index(vmx->host_state.gs_sel);
1450 #else
1451                 loadsegment(gs, vmx->host_state.gs_sel);
1452 #endif
1453         }
1454         if (vmx->host_state.fs_reload_needed)
1455                 loadsegment(fs, vmx->host_state.fs_sel);
1456         reload_tss();
1457 #ifdef CONFIG_X86_64
1458         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1459 #endif
1460         if (current_thread_info()->status & TS_USEDFPU)
1461                 clts();
1462         load_gdt(&__get_cpu_var(host_gdt));
1463 }
1464
1465 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1466 {
1467         preempt_disable();
1468         __vmx_load_host_state(vmx);
1469         preempt_enable();
1470 }
1471
1472 /*
1473  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1474  * vcpu mutex is already taken.
1475  */
1476 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1477 {
1478         struct vcpu_vmx *vmx = to_vmx(vcpu);
1479         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1480
1481         if (!vmm_exclusive)
1482                 kvm_cpu_vmxon(phys_addr);
1483         else if (vmx->loaded_vmcs->cpu != cpu)
1484                 loaded_vmcs_clear(vmx->loaded_vmcs);
1485
1486         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1487                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1488                 vmcs_load(vmx->loaded_vmcs->vmcs);
1489         }
1490
1491         if (vmx->loaded_vmcs->cpu != cpu) {
1492                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1493                 unsigned long sysenter_esp;
1494
1495                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1496                 local_irq_disable();
1497                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1498                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1499                 local_irq_enable();
1500
1501                 /*
1502                  * Linux uses per-cpu TSS and GDT, so set these when switching
1503                  * processors.
1504                  */
1505                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1506                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1507
1508                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1509                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1510                 vmx->loaded_vmcs->cpu = cpu;
1511         }
1512 }
1513
1514 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1515 {
1516         __vmx_load_host_state(to_vmx(vcpu));
1517         if (!vmm_exclusive) {
1518                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1519                 vcpu->cpu = -1;
1520                 kvm_cpu_vmxoff();
1521         }
1522 }
1523
1524 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1525 {
1526         ulong cr0;
1527
1528         if (vcpu->fpu_active)
1529                 return;
1530         vcpu->fpu_active = 1;
1531         cr0 = vmcs_readl(GUEST_CR0);
1532         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1533         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1534         vmcs_writel(GUEST_CR0, cr0);
1535         update_exception_bitmap(vcpu);
1536         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1537         if (is_guest_mode(vcpu))
1538                 vcpu->arch.cr0_guest_owned_bits &=
1539                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1540         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1541 }
1542
1543 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1544
1545 /*
1546  * Return the cr0 value that a nested guest would read. This is a combination
1547  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1548  * its hypervisor (cr0_read_shadow).
1549  */
1550 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1551 {
1552         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1553                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1554 }
1555 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1556 {
1557         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1558                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1559 }
1560
1561 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1562 {
1563         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1564          * set this *before* calling this function.
1565          */
1566         vmx_decache_cr0_guest_bits(vcpu);
1567         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1568         update_exception_bitmap(vcpu);
1569         vcpu->arch.cr0_guest_owned_bits = 0;
1570         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1571         if (is_guest_mode(vcpu)) {
1572                 /*
1573                  * L1's specified read shadow might not contain the TS bit,
1574                  * so now that we turned on shadowing of this bit, we need to
1575                  * set this bit of the shadow. Like in nested_vmx_run we need
1576                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1577                  * up-to-date here because we just decached cr0.TS (and we'll
1578                  * only update vmcs12->guest_cr0 on nested exit).
1579                  */
1580                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1581                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1582                         (vcpu->arch.cr0 & X86_CR0_TS);
1583                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1584         } else
1585                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1586 }
1587
1588 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1589 {
1590         unsigned long rflags, save_rflags;
1591
1592         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1593                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1594                 rflags = vmcs_readl(GUEST_RFLAGS);
1595                 if (to_vmx(vcpu)->rmode.vm86_active) {
1596                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1597                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1598                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1599                 }
1600                 to_vmx(vcpu)->rflags = rflags;
1601         }
1602         return to_vmx(vcpu)->rflags;
1603 }
1604
1605 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1606 {
1607         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1608         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1609         to_vmx(vcpu)->rflags = rflags;
1610         if (to_vmx(vcpu)->rmode.vm86_active) {
1611                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1612                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1613         }
1614         vmcs_writel(GUEST_RFLAGS, rflags);
1615 }
1616
1617 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1618 {
1619         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1620         int ret = 0;
1621
1622         if (interruptibility & GUEST_INTR_STATE_STI)
1623                 ret |= KVM_X86_SHADOW_INT_STI;
1624         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1625                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1626
1627         return ret & mask;
1628 }
1629
1630 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1631 {
1632         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1633         u32 interruptibility = interruptibility_old;
1634
1635         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1636
1637         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1638                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1639         else if (mask & KVM_X86_SHADOW_INT_STI)
1640                 interruptibility |= GUEST_INTR_STATE_STI;
1641
1642         if ((interruptibility != interruptibility_old))
1643                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1644 }
1645
1646 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1647 {
1648         unsigned long rip;
1649
1650         rip = kvm_rip_read(vcpu);
1651         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1652         kvm_rip_write(vcpu, rip);
1653
1654         /* skipping an emulated instruction also counts */
1655         vmx_set_interrupt_shadow(vcpu, 0);
1656 }
1657
1658 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1659 {
1660         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1661          * explicitly skip the instruction because if the HLT state is set, then
1662          * the instruction is already executing and RIP has already been
1663          * advanced. */
1664         if (!yield_on_hlt &&
1665             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1666                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1667 }
1668
1669 /*
1670  * KVM wants to inject page-faults which it got to the guest. This function
1671  * checks whether in a nested guest, we need to inject them to L1 or L2.
1672  * This function assumes it is called with the exit reason in vmcs02 being
1673  * a #PF exception (this is the only case in which KVM injects a #PF when L2
1674  * is running).
1675  */
1676 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1677 {
1678         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1679
1680         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1681         if (!(vmcs12->exception_bitmap & PF_VECTOR))
1682                 return 0;
1683
1684         nested_vmx_vmexit(vcpu);
1685         return 1;
1686 }
1687
1688 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1689                                 bool has_error_code, u32 error_code,
1690                                 bool reinject)
1691 {
1692         struct vcpu_vmx *vmx = to_vmx(vcpu);
1693         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1694
1695         if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1696                 nested_pf_handled(vcpu))
1697                 return;
1698
1699         if (has_error_code) {
1700                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1701                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1702         }
1703
1704         if (vmx->rmode.vm86_active) {
1705                 int inc_eip = 0;
1706                 if (kvm_exception_is_soft(nr))
1707                         inc_eip = vcpu->arch.event_exit_inst_len;
1708                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1709                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1710                 return;
1711         }
1712
1713         if (kvm_exception_is_soft(nr)) {
1714                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1715                              vmx->vcpu.arch.event_exit_inst_len);
1716                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1717         } else
1718                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1719
1720         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1721         vmx_clear_hlt(vcpu);
1722 }
1723
1724 static bool vmx_rdtscp_supported(void)
1725 {
1726         return cpu_has_vmx_rdtscp();
1727 }
1728
1729 /*
1730  * Swap MSR entry in host/guest MSR entry array.
1731  */
1732 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1733 {
1734         struct shared_msr_entry tmp;
1735
1736         tmp = vmx->guest_msrs[to];
1737         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1738         vmx->guest_msrs[from] = tmp;
1739 }
1740
1741 /*
1742  * Set up the vmcs to automatically save and restore system
1743  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1744  * mode, as fiddling with msrs is very expensive.
1745  */
1746 static void setup_msrs(struct vcpu_vmx *vmx)
1747 {
1748         int save_nmsrs, index;
1749         unsigned long *msr_bitmap;
1750
1751         save_nmsrs = 0;
1752 #ifdef CONFIG_X86_64
1753         if (is_long_mode(&vmx->vcpu)) {
1754                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1755                 if (index >= 0)
1756                         move_msr_up(vmx, index, save_nmsrs++);
1757                 index = __find_msr_index(vmx, MSR_LSTAR);
1758                 if (index >= 0)
1759                         move_msr_up(vmx, index, save_nmsrs++);
1760                 index = __find_msr_index(vmx, MSR_CSTAR);
1761                 if (index >= 0)
1762                         move_msr_up(vmx, index, save_nmsrs++);
1763                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1764                 if (index >= 0 && vmx->rdtscp_enabled)
1765                         move_msr_up(vmx, index, save_nmsrs++);
1766                 /*
1767                  * MSR_STAR is only needed on long mode guests, and only
1768                  * if efer.sce is enabled.
1769                  */
1770                 index = __find_msr_index(vmx, MSR_STAR);
1771                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1772                         move_msr_up(vmx, index, save_nmsrs++);
1773         }
1774 #endif
1775         index = __find_msr_index(vmx, MSR_EFER);
1776         if (index >= 0 && update_transition_efer(vmx, index))
1777                 move_msr_up(vmx, index, save_nmsrs++);
1778
1779         vmx->save_nmsrs = save_nmsrs;
1780
1781         if (cpu_has_vmx_msr_bitmap()) {
1782                 if (is_long_mode(&vmx->vcpu))
1783                         msr_bitmap = vmx_msr_bitmap_longmode;
1784                 else
1785                         msr_bitmap = vmx_msr_bitmap_legacy;
1786
1787                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1788         }
1789 }
1790
1791 /*
1792  * reads and returns guest's timestamp counter "register"
1793  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1794  */
1795 static u64 guest_read_tsc(void)
1796 {
1797         u64 host_tsc, tsc_offset;
1798
1799         rdtscll(host_tsc);
1800         tsc_offset = vmcs_read64(TSC_OFFSET);
1801         return host_tsc + tsc_offset;
1802 }
1803
1804 /*
1805  * Like guest_read_tsc, but always returns L1's notion of the timestamp
1806  * counter, even if a nested guest (L2) is currently running.
1807  */
1808 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1809 {
1810         u64 host_tsc, tsc_offset;
1811
1812         rdtscll(host_tsc);
1813         tsc_offset = is_guest_mode(vcpu) ?
1814                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1815                 vmcs_read64(TSC_OFFSET);
1816         return host_tsc + tsc_offset;
1817 }
1818
1819 /*
1820  * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1821  * ioctl. In this case the call-back should update internal vmx state to make
1822  * the changes effective.
1823  */
1824 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1825 {
1826         /* Nothing to do here */
1827 }
1828
1829 /*
1830  * writes 'offset' into guest's timestamp counter offset register
1831  */
1832 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1833 {
1834         if (is_guest_mode(vcpu)) {
1835                 /*
1836                  * We're here if L1 chose not to trap WRMSR to TSC. According
1837                  * to the spec, this should set L1's TSC; The offset that L1
1838                  * set for L2 remains unchanged, and still needs to be added
1839                  * to the newly set TSC to get L2's TSC.
1840                  */
1841                 struct vmcs12 *vmcs12;
1842                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1843                 /* recalculate vmcs02.TSC_OFFSET: */
1844                 vmcs12 = get_vmcs12(vcpu);
1845                 vmcs_write64(TSC_OFFSET, offset +
1846                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1847                          vmcs12->tsc_offset : 0));
1848         } else {
1849                 vmcs_write64(TSC_OFFSET, offset);
1850         }
1851 }
1852
1853 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1854 {
1855         u64 offset = vmcs_read64(TSC_OFFSET);
1856         vmcs_write64(TSC_OFFSET, offset + adjustment);
1857         if (is_guest_mode(vcpu)) {
1858                 /* Even when running L2, the adjustment needs to apply to L1 */
1859                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1860         }
1861 }
1862
1863 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1864 {
1865         return target_tsc - native_read_tsc();
1866 }
1867
1868 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1869 {
1870         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1871         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1872 }
1873
1874 /*
1875  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1876  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1877  * all guests if the "nested" module option is off, and can also be disabled
1878  * for a single guest by disabling its VMX cpuid bit.
1879  */
1880 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1881 {
1882         return nested && guest_cpuid_has_vmx(vcpu);
1883 }
1884
1885 /*
1886  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1887  * returned for the various VMX controls MSRs when nested VMX is enabled.
1888  * The same values should also be used to verify that vmcs12 control fields are
1889  * valid during nested entry from L1 to L2.
1890  * Each of these control msrs has a low and high 32-bit half: A low bit is on
1891  * if the corresponding bit in the (32-bit) control field *must* be on, and a
1892  * bit in the high half is on if the corresponding bit in the control field
1893  * may be on. See also vmx_control_verify().
1894  * TODO: allow these variables to be modified (downgraded) by module options
1895  * or other means.
1896  */
1897 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1898 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1899 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1900 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1901 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1902 static __init void nested_vmx_setup_ctls_msrs(void)
1903 {
1904         /*
1905          * Note that as a general rule, the high half of the MSRs (bits in
1906          * the control fields which may be 1) should be initialized by the
1907          * intersection of the underlying hardware's MSR (i.e., features which
1908          * can be supported) and the list of features we want to expose -
1909          * because they are known to be properly supported in our code.
1910          * Also, usually, the low half of the MSRs (bits which must be 1) can
1911          * be set to 0, meaning that L1 may turn off any of these bits. The
1912          * reason is that if one of these bits is necessary, it will appear
1913          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1914          * fields of vmcs01 and vmcs02, will turn these bits off - and
1915          * nested_vmx_exit_handled() will not pass related exits to L1.
1916          * These rules have exceptions below.
1917          */
1918
1919         /* pin-based controls */
1920         /*
1921          * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1922          * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1923          */
1924         nested_vmx_pinbased_ctls_low = 0x16 ;
1925         nested_vmx_pinbased_ctls_high = 0x16 |
1926                 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1927                 PIN_BASED_VIRTUAL_NMIS;
1928
1929         /* exit controls */
1930         nested_vmx_exit_ctls_low = 0;
1931         /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1932 #ifdef CONFIG_X86_64
1933         nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1934 #else
1935         nested_vmx_exit_ctls_high = 0;
1936 #endif
1937
1938         /* entry controls */
1939         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1940                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1941         nested_vmx_entry_ctls_low = 0;
1942         nested_vmx_entry_ctls_high &=
1943                 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1944
1945         /* cpu-based controls */
1946         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1947                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1948         nested_vmx_procbased_ctls_low = 0;
1949         nested_vmx_procbased_ctls_high &=
1950                 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1951                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1952                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1953                 CPU_BASED_CR3_STORE_EXITING |
1954 #ifdef CONFIG_X86_64
1955                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1956 #endif
1957                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1958                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1959                 CPU_BASED_RDPMC_EXITING |
1960                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1961         /*
1962          * We can allow some features even when not supported by the
1963          * hardware. For example, L1 can specify an MSR bitmap - and we
1964          * can use it to avoid exits to L1 - even when L0 runs L2
1965          * without MSR bitmaps.
1966          */
1967         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1968
1969         /* secondary cpu-based controls */
1970         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1971                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1972         nested_vmx_secondary_ctls_low = 0;
1973         nested_vmx_secondary_ctls_high &=
1974                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1975 }
1976
1977 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1978 {
1979         /*
1980          * Bits 0 in high must be 0, and bits 1 in low must be 1.
1981          */
1982         return ((control & high) | low) == control;
1983 }
1984
1985 static inline u64 vmx_control_msr(u32 low, u32 high)
1986 {
1987         return low | ((u64)high << 32);
1988 }
1989
1990 /*
1991  * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
1992  * also let it use VMX-specific MSRs.
1993  * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
1994  * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
1995  * like all other MSRs).
1996  */
1997 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1998 {
1999         if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2000                      msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2001                 /*
2002                  * According to the spec, processors which do not support VMX
2003                  * should throw a #GP(0) when VMX capability MSRs are read.
2004                  */
2005                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2006                 return 1;
2007         }
2008
2009         switch (msr_index) {
2010         case MSR_IA32_FEATURE_CONTROL:
2011                 *pdata = 0;
2012                 break;
2013         case MSR_IA32_VMX_BASIC:
2014                 /*
2015                  * This MSR reports some information about VMX support. We
2016                  * should return information about the VMX we emulate for the
2017                  * guest, and the VMCS structure we give it - not about the
2018                  * VMX support of the underlying hardware.
2019                  */
2020                 *pdata = VMCS12_REVISION |
2021                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2022                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2023                 break;
2024         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2025         case MSR_IA32_VMX_PINBASED_CTLS:
2026                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2027                                         nested_vmx_pinbased_ctls_high);
2028                 break;
2029         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2030         case MSR_IA32_VMX_PROCBASED_CTLS:
2031                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2032                                         nested_vmx_procbased_ctls_high);
2033                 break;
2034         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2035         case MSR_IA32_VMX_EXIT_CTLS:
2036                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2037                                         nested_vmx_exit_ctls_high);
2038                 break;
2039         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2040         case MSR_IA32_VMX_ENTRY_CTLS:
2041                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2042                                         nested_vmx_entry_ctls_high);
2043                 break;
2044         case MSR_IA32_VMX_MISC:
2045                 *pdata = 0;
2046                 break;
2047         /*
2048          * These MSRs specify bits which the guest must keep fixed (on or off)
2049          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2050          * We picked the standard core2 setting.
2051          */
2052 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2053 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2054         case MSR_IA32_VMX_CR0_FIXED0:
2055                 *pdata = VMXON_CR0_ALWAYSON;
2056                 break;
2057         case MSR_IA32_VMX_CR0_FIXED1:
2058                 *pdata = -1ULL;
2059                 break;
2060         case MSR_IA32_VMX_CR4_FIXED0:
2061                 *pdata = VMXON_CR4_ALWAYSON;
2062                 break;
2063         case MSR_IA32_VMX_CR4_FIXED1:
2064                 *pdata = -1ULL;
2065                 break;
2066         case MSR_IA32_VMX_VMCS_ENUM:
2067                 *pdata = 0x1f;
2068                 break;
2069         case MSR_IA32_VMX_PROCBASED_CTLS2:
2070                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2071                                         nested_vmx_secondary_ctls_high);
2072                 break;
2073         case MSR_IA32_VMX_EPT_VPID_CAP:
2074                 /* Currently, no nested ept or nested vpid */
2075                 *pdata = 0;
2076                 break;
2077         default:
2078                 return 0;
2079         }
2080
2081         return 1;
2082 }
2083
2084 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2085 {
2086         if (!nested_vmx_allowed(vcpu))
2087                 return 0;
2088
2089         if (msr_index == MSR_IA32_FEATURE_CONTROL)
2090                 /* TODO: the right thing. */
2091                 return 1;
2092         /*
2093          * No need to treat VMX capability MSRs specially: If we don't handle
2094          * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2095          */
2096         return 0;
2097 }
2098
2099 /*
2100  * Reads an msr value (of 'msr_index') into 'pdata'.
2101  * Returns 0 on success, non-0 otherwise.
2102  * Assumes vcpu_load() was already called.
2103  */
2104 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2105 {
2106         u64 data;
2107         struct shared_msr_entry *msr;
2108
2109         if (!pdata) {
2110                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2111                 return -EINVAL;
2112         }
2113
2114         switch (msr_index) {
2115 #ifdef CONFIG_X86_64
2116         case MSR_FS_BASE:
2117                 data = vmcs_readl(GUEST_FS_BASE);
2118                 break;
2119         case MSR_GS_BASE:
2120                 data = vmcs_readl(GUEST_GS_BASE);
2121                 break;
2122         case MSR_KERNEL_GS_BASE:
2123                 vmx_load_host_state(to_vmx(vcpu));
2124                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2125                 break;
2126 #endif
2127         case MSR_EFER:
2128                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2129         case MSR_IA32_TSC:
2130                 data = guest_read_tsc();
2131                 break;
2132         case MSR_IA32_SYSENTER_CS:
2133                 data = vmcs_read32(GUEST_SYSENTER_CS);
2134                 break;
2135         case MSR_IA32_SYSENTER_EIP:
2136                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2137                 break;
2138         case MSR_IA32_SYSENTER_ESP:
2139                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2140                 break;
2141         case MSR_TSC_AUX:
2142                 if (!to_vmx(vcpu)->rdtscp_enabled)
2143                         return 1;
2144                 /* Otherwise falls through */
2145         default:
2146                 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2147                         return 0;
2148                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2149                 if (msr) {
2150                         data = msr->data;
2151                         break;
2152                 }
2153                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2154         }
2155
2156         *pdata = data;
2157         return 0;
2158 }
2159
2160 /*
2161  * Writes msr value into into the appropriate "register".
2162  * Returns 0 on success, non-0 otherwise.
2163  * Assumes vcpu_load() was already called.
2164  */
2165 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2166 {
2167         struct vcpu_vmx *vmx = to_vmx(vcpu);
2168         struct shared_msr_entry *msr;
2169         int ret = 0;
2170
2171         switch (msr_index) {
2172         case MSR_EFER:
2173                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2174                 break;
2175 #ifdef CONFIG_X86_64
2176         case MSR_FS_BASE:
2177                 vmx_segment_cache_clear(vmx);
2178                 vmcs_writel(GUEST_FS_BASE, data);
2179                 break;
2180         case MSR_GS_BASE:
2181                 vmx_segment_cache_clear(vmx);
2182                 vmcs_writel(GUEST_GS_BASE, data);
2183                 break;
2184         case MSR_KERNEL_GS_BASE:
2185                 vmx_load_host_state(vmx);
2186                 vmx->msr_guest_kernel_gs_base = data;
2187                 break;
2188 #endif
2189         case MSR_IA32_SYSENTER_CS:
2190                 vmcs_write32(GUEST_SYSENTER_CS, data);
2191                 break;
2192         case MSR_IA32_SYSENTER_EIP:
2193                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2194                 break;
2195         case MSR_IA32_SYSENTER_ESP:
2196                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2197                 break;
2198         case MSR_IA32_TSC:
2199                 kvm_write_tsc(vcpu, data);
2200                 break;
2201         case MSR_IA32_CR_PAT:
2202                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2203                         vmcs_write64(GUEST_IA32_PAT, data);
2204                         vcpu->arch.pat = data;
2205                         break;
2206                 }
2207                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2208                 break;
2209         case MSR_TSC_AUX:
2210                 if (!vmx->rdtscp_enabled)
2211                         return 1;
2212                 /* Check reserved bit, higher 32 bits should be zero */
2213                 if ((data >> 32) != 0)
2214                         return 1;
2215                 /* Otherwise falls through */
2216         default:
2217                 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2218                         break;
2219                 msr = find_msr_entry(vmx, msr_index);
2220                 if (msr) {
2221                         msr->data = data;
2222                         break;
2223                 }
2224                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2225         }
2226
2227         return ret;
2228 }
2229
2230 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2231 {
2232         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2233         switch (reg) {
2234         case VCPU_REGS_RSP:
2235                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2236                 break;
2237         case VCPU_REGS_RIP:
2238                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2239                 break;
2240         case VCPU_EXREG_PDPTR:
2241                 if (enable_ept)
2242                         ept_save_pdptrs(vcpu);
2243                 break;
2244         default:
2245                 break;
2246         }
2247 }
2248
2249 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2250 {
2251         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2252                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2253         else
2254                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2255
2256         update_exception_bitmap(vcpu);
2257 }
2258
2259 static __init int cpu_has_kvm_support(void)
2260 {
2261         return cpu_has_vmx();
2262 }
2263
2264 static __init int vmx_disabled_by_bios(void)
2265 {
2266         u64 msr;
2267
2268         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2269         if (msr & FEATURE_CONTROL_LOCKED) {
2270                 /* launched w/ TXT and VMX disabled */
2271                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2272                         && tboot_enabled())
2273                         return 1;
2274                 /* launched w/o TXT and VMX only enabled w/ TXT */
2275                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2276                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2277                         && !tboot_enabled()) {
2278                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2279                                 "activate TXT before enabling KVM\n");
2280                         return 1;
2281                 }
2282                 /* launched w/o TXT and VMX disabled */
2283                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2284                         && !tboot_enabled())
2285                         return 1;
2286         }
2287
2288         return 0;
2289 }
2290
2291 static void kvm_cpu_vmxon(u64 addr)
2292 {
2293         asm volatile (ASM_VMX_VMXON_RAX
2294                         : : "a"(&addr), "m"(addr)
2295                         : "memory", "cc");
2296 }
2297
2298 static int hardware_enable(void *garbage)
2299 {
2300         int cpu = raw_smp_processor_id();
2301         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2302         u64 old, test_bits;
2303
2304         if (read_cr4() & X86_CR4_VMXE)
2305                 return -EBUSY;
2306
2307         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2308         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2309
2310         test_bits = FEATURE_CONTROL_LOCKED;
2311         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2312         if (tboot_enabled())
2313                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2314
2315         if ((old & test_bits) != test_bits) {
2316                 /* enable and lock */
2317                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2318         }
2319         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2320
2321         if (vmm_exclusive) {
2322                 kvm_cpu_vmxon(phys_addr);
2323                 ept_sync_global();
2324         }
2325
2326         store_gdt(&__get_cpu_var(host_gdt));
2327
2328         return 0;
2329 }
2330
2331 static void vmclear_local_loaded_vmcss(void)
2332 {
2333         int cpu = raw_smp_processor_id();
2334         struct loaded_vmcs *v, *n;
2335
2336         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2337                                  loaded_vmcss_on_cpu_link)
2338                 __loaded_vmcs_clear(v);
2339 }
2340
2341
2342 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2343  * tricks.
2344  */
2345 static void kvm_cpu_vmxoff(void)
2346 {
2347         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2348 }
2349
2350 static void hardware_disable(void *garbage)
2351 {
2352         if (vmm_exclusive) {
2353                 vmclear_local_loaded_vmcss();
2354                 kvm_cpu_vmxoff();
2355         }
2356         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2357 }
2358
2359 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2360                                       u32 msr, u32 *result)
2361 {
2362         u32 vmx_msr_low, vmx_msr_high;
2363         u32 ctl = ctl_min | ctl_opt;
2364
2365         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2366
2367         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2368         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2369
2370         /* Ensure minimum (required) set of control bits are supported. */
2371         if (ctl_min & ~ctl)
2372                 return -EIO;
2373
2374         *result = ctl;
2375         return 0;
2376 }
2377
2378 static __init bool allow_1_setting(u32 msr, u32 ctl)
2379 {
2380         u32 vmx_msr_low, vmx_msr_high;
2381
2382         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2383         return vmx_msr_high & ctl;
2384 }
2385
2386 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2387 {
2388         u32 vmx_msr_low, vmx_msr_high;
2389         u32 min, opt, min2, opt2;
2390         u32 _pin_based_exec_control = 0;
2391         u32 _cpu_based_exec_control = 0;
2392         u32 _cpu_based_2nd_exec_control = 0;
2393         u32 _vmexit_control = 0;
2394         u32 _vmentry_control = 0;
2395
2396         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2397         opt = PIN_BASED_VIRTUAL_NMIS;
2398         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2399                                 &_pin_based_exec_control) < 0)
2400                 return -EIO;
2401
2402         min =
2403 #ifdef CONFIG_X86_64
2404               CPU_BASED_CR8_LOAD_EXITING |
2405               CPU_BASED_CR8_STORE_EXITING |
2406 #endif
2407               CPU_BASED_CR3_LOAD_EXITING |
2408               CPU_BASED_CR3_STORE_EXITING |
2409               CPU_BASED_USE_IO_BITMAPS |
2410               CPU_BASED_MOV_DR_EXITING |
2411               CPU_BASED_USE_TSC_OFFSETING |
2412               CPU_BASED_MWAIT_EXITING |
2413               CPU_BASED_MONITOR_EXITING |
2414               CPU_BASED_INVLPG_EXITING |
2415               CPU_BASED_RDPMC_EXITING;
2416
2417         if (yield_on_hlt)
2418                 min |= CPU_BASED_HLT_EXITING;
2419
2420         opt = CPU_BASED_TPR_SHADOW |
2421               CPU_BASED_USE_MSR_BITMAPS |
2422               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2423         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2424                                 &_cpu_based_exec_control) < 0)
2425                 return -EIO;
2426 #ifdef CONFIG_X86_64
2427         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2428                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2429                                            ~CPU_BASED_CR8_STORE_EXITING;
2430 #endif
2431         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2432                 min2 = 0;
2433                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2434                         SECONDARY_EXEC_WBINVD_EXITING |
2435                         SECONDARY_EXEC_ENABLE_VPID |
2436                         SECONDARY_EXEC_ENABLE_EPT |
2437                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2438                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2439                         SECONDARY_EXEC_RDTSCP;
2440                 if (adjust_vmx_controls(min2, opt2,
2441                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2442                                         &_cpu_based_2nd_exec_control) < 0)
2443                         return -EIO;
2444         }
2445 #ifndef CONFIG_X86_64
2446         if (!(_cpu_based_2nd_exec_control &
2447                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2448                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2449 #endif
2450         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2451                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2452                    enabled */
2453                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2454                                              CPU_BASED_CR3_STORE_EXITING |
2455                                              CPU_BASED_INVLPG_EXITING);
2456                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2457                       vmx_capability.ept, vmx_capability.vpid);
2458         }
2459
2460         min = 0;
2461 #ifdef CONFIG_X86_64
2462         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2463 #endif
2464         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2465         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2466                                 &_vmexit_control) < 0)
2467                 return -EIO;
2468
2469         min = 0;
2470         opt = VM_ENTRY_LOAD_IA32_PAT;
2471         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2472                                 &_vmentry_control) < 0)
2473                 return -EIO;
2474
2475         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2476
2477         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2478         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2479                 return -EIO;
2480
2481 #ifdef CONFIG_X86_64
2482         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2483         if (vmx_msr_high & (1u<<16))
2484                 return -EIO;
2485 #endif
2486
2487         /* Require Write-Back (WB) memory type for VMCS accesses. */
2488         if (((vmx_msr_high >> 18) & 15) != 6)
2489                 return -EIO;
2490
2491         vmcs_conf->size = vmx_msr_high & 0x1fff;
2492         vmcs_conf->order = get_order(vmcs_config.size);
2493         vmcs_conf->revision_id = vmx_msr_low;
2494
2495         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2496         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2497         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2498         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2499         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2500
2501         cpu_has_load_ia32_efer =
2502                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2503                                 VM_ENTRY_LOAD_IA32_EFER)
2504                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2505                                    VM_EXIT_LOAD_IA32_EFER);
2506
2507         cpu_has_load_perf_global_ctrl =
2508                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2509                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2510                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2511                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2512
2513         /*
2514          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2515          * but due to arrata below it can't be used. Workaround is to use
2516          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2517          *
2518          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2519          *
2520          * AAK155             (model 26)
2521          * AAP115             (model 30)
2522          * AAT100             (model 37)
2523          * BC86,AAY89,BD102   (model 44)
2524          * BA97               (model 46)
2525          *
2526          */
2527         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2528                 switch (boot_cpu_data.x86_model) {
2529                 case 26:
2530                 case 30:
2531                 case 37:
2532                 case 44:
2533                 case 46:
2534                         cpu_has_load_perf_global_ctrl = false;
2535                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2536                                         "does not work properly. Using workaround\n");
2537                         break;
2538                 default:
2539                         break;
2540                 }
2541         }
2542
2543         return 0;
2544 }
2545
2546 static struct vmcs *alloc_vmcs_cpu(int cpu)
2547 {
2548         int node = cpu_to_node(cpu);
2549         struct page *pages;
2550         struct vmcs *vmcs;
2551
2552         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2553         if (!pages)
2554                 return NULL;
2555         vmcs = page_address(pages);
2556         memset(vmcs, 0, vmcs_config.size);
2557         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2558         return vmcs;
2559 }
2560
2561 static struct vmcs *alloc_vmcs(void)
2562 {
2563         return alloc_vmcs_cpu(raw_smp_processor_id());
2564 }
2565
2566 static void free_vmcs(struct vmcs *vmcs)
2567 {
2568         free_pages((unsigned long)vmcs, vmcs_config.order);
2569 }
2570
2571 /*
2572  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2573  */
2574 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2575 {
2576         if (!loaded_vmcs->vmcs)
2577                 return;
2578         loaded_vmcs_clear(loaded_vmcs);
2579         free_vmcs(loaded_vmcs->vmcs);
2580         loaded_vmcs->vmcs = NULL;
2581 }
2582
2583 static void free_kvm_area(void)
2584 {
2585         int cpu;
2586
2587         for_each_possible_cpu(cpu) {
2588                 free_vmcs(per_cpu(vmxarea, cpu));
2589                 per_cpu(vmxarea, cpu) = NULL;
2590         }
2591 }
2592
2593 static __init int alloc_kvm_area(void)
2594 {
2595         int cpu;
2596
2597         for_each_possible_cpu(cpu) {
2598                 struct vmcs *vmcs;
2599
2600                 vmcs = alloc_vmcs_cpu(cpu);
2601                 if (!vmcs) {
2602                         free_kvm_area();
2603                         return -ENOMEM;
2604                 }
2605
2606                 per_cpu(vmxarea, cpu) = vmcs;
2607         }
2608         return 0;
2609 }
2610
2611 static __init int hardware_setup(void)
2612 {
2613         if (setup_vmcs_config(&vmcs_config) < 0)
2614                 return -EIO;
2615
2616         if (boot_cpu_has(X86_FEATURE_NX))
2617                 kvm_enable_efer_bits(EFER_NX);
2618
2619         if (!cpu_has_vmx_vpid())
2620                 enable_vpid = 0;
2621
2622         if (!cpu_has_vmx_ept() ||
2623             !cpu_has_vmx_ept_4levels()) {
2624                 enable_ept = 0;
2625                 enable_unrestricted_guest = 0;
2626         }
2627
2628         if (!cpu_has_vmx_unrestricted_guest())
2629                 enable_unrestricted_guest = 0;
2630
2631         if (!cpu_has_vmx_flexpriority())
2632                 flexpriority_enabled = 0;
2633
2634         if (!cpu_has_vmx_tpr_shadow())
2635                 kvm_x86_ops->update_cr8_intercept = NULL;
2636
2637         if (enable_ept && !cpu_has_vmx_ept_2m_page())
2638                 kvm_disable_largepages();
2639
2640         if (!cpu_has_vmx_ple())
2641                 ple_gap = 0;
2642
2643         if (nested)
2644                 nested_vmx_setup_ctls_msrs();
2645
2646         return alloc_kvm_area();
2647 }
2648
2649 static __exit void hardware_unsetup(void)
2650 {
2651         free_kvm_area();
2652 }
2653
2654 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2655 {
2656         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2657
2658         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2659                 vmcs_write16(sf->selector, save->selector);
2660                 vmcs_writel(sf->base, save->base);
2661                 vmcs_write32(sf->limit, save->limit);
2662                 vmcs_write32(sf->ar_bytes, save->ar);
2663         } else {
2664                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2665                         << AR_DPL_SHIFT;
2666                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2667         }
2668 }
2669
2670 static void enter_pmode(struct kvm_vcpu *vcpu)
2671 {
2672         unsigned long flags;
2673         struct vcpu_vmx *vmx = to_vmx(vcpu);
2674
2675         vmx->emulation_required = 1;
2676         vmx->rmode.vm86_active = 0;
2677
2678         vmx_segment_cache_clear(vmx);
2679
2680         vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2681         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2682         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2683         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2684
2685         flags = vmcs_readl(GUEST_RFLAGS);
2686         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2687         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2688         vmcs_writel(GUEST_RFLAGS, flags);
2689
2690         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2691                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2692
2693         update_exception_bitmap(vcpu);
2694
2695         if (emulate_invalid_guest_state)
2696                 return;
2697
2698         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2699         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2700         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2701         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2702
2703         vmx_segment_cache_clear(vmx);
2704
2705         vmcs_write16(GUEST_SS_SELECTOR, 0);
2706         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2707
2708         vmcs_write16(GUEST_CS_SELECTOR,
2709                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2710         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2711 }
2712
2713 static gva_t rmode_tss_base(struct kvm *kvm)
2714 {
2715         if (!kvm->arch.tss_addr) {
2716                 struct kvm_memslots *slots;
2717                 struct kvm_memory_slot *slot;
2718                 gfn_t base_gfn;
2719
2720                 slots = kvm_memslots(kvm);
2721                 slot = id_to_memslot(slots, 0);
2722                 base_gfn = slot->base_gfn + slot->npages - 3;
2723
2724                 return base_gfn << PAGE_SHIFT;
2725         }
2726         return kvm->arch.tss_addr;
2727 }
2728
2729 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2730 {
2731         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2732
2733         save->selector = vmcs_read16(sf->selector);
2734         save->base = vmcs_readl(sf->base);
2735         save->limit = vmcs_read32(sf->limit);
2736         save->ar = vmcs_read32(sf->ar_bytes);
2737         vmcs_write16(sf->selector, save->base >> 4);
2738         vmcs_write32(sf->base, save->base & 0xffff0);
2739         vmcs_write32(sf->limit, 0xffff);
2740         vmcs_write32(sf->ar_bytes, 0xf3);
2741         if (save->base & 0xf)
2742                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2743                             " aligned when entering protected mode (seg=%d)",
2744                             seg);
2745 }
2746
2747 static void enter_rmode(struct kvm_vcpu *vcpu)
2748 {
2749         unsigned long flags;
2750         struct vcpu_vmx *vmx = to_vmx(vcpu);
2751
2752         if (enable_unrestricted_guest)
2753                 return;
2754
2755         vmx->emulation_required = 1;
2756         vmx->rmode.vm86_active = 1;
2757
2758         /*
2759          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2760          * vcpu. Call it here with phys address pointing 16M below 4G.
2761          */
2762         if (!vcpu->kvm->arch.tss_addr) {
2763                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2764                              "called before entering vcpu\n");
2765                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2766                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2767                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2768         }
2769
2770         vmx_segment_cache_clear(vmx);
2771
2772         vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2773         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2774         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2775
2776         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2777         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2778
2779         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2780         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2781
2782         flags = vmcs_readl(GUEST_RFLAGS);
2783         vmx->rmode.save_rflags = flags;
2784
2785         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2786
2787         vmcs_writel(GUEST_RFLAGS, flags);
2788         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2789         update_exception_bitmap(vcpu);
2790
2791         if (emulate_invalid_guest_state)
2792                 goto continue_rmode;
2793
2794         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
2795         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
2796         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
2797
2798         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
2799         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2800         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
2801                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
2802         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
2803
2804         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
2805         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
2806         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
2807         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
2808
2809 continue_rmode:
2810         kvm_mmu_reset_context(vcpu);
2811 }
2812
2813 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2814 {
2815         struct vcpu_vmx *vmx = to_vmx(vcpu);
2816         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2817
2818         if (!msr)
2819                 return;
2820
2821         /*
2822          * Force kernel_gs_base reloading before EFER changes, as control
2823          * of this msr depends on is_long_mode().
2824          */
2825         vmx_load_host_state(to_vmx(vcpu));
2826         vcpu->arch.efer = efer;
2827         if (efer & EFER_LMA) {
2828                 vmcs_write32(VM_ENTRY_CONTROLS,
2829                              vmcs_read32(VM_ENTRY_CONTROLS) |
2830                              VM_ENTRY_IA32E_MODE);
2831                 msr->data = efer;
2832         } else {
2833                 vmcs_write32(VM_ENTRY_CONTROLS,
2834                              vmcs_read32(VM_ENTRY_CONTROLS) &
2835                              ~VM_ENTRY_IA32E_MODE);
2836
2837                 msr->data = efer & ~EFER_LME;
2838         }
2839         setup_msrs(vmx);
2840 }
2841
2842 #ifdef CONFIG_X86_64
2843
2844 static void enter_lmode(struct kvm_vcpu *vcpu)
2845 {
2846         u32 guest_tr_ar;
2847
2848         vmx_segment_cache_clear(to_vmx(vcpu));
2849
2850         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2851         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2852                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2853                                      __func__);
2854                 vmcs_write32(GUEST_TR_AR_BYTES,
2855                              (guest_tr_ar & ~AR_TYPE_MASK)
2856                              | AR_TYPE_BUSY_64_TSS);
2857         }
2858         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2859 }
2860
2861 static void exit_lmode(struct kvm_vcpu *vcpu)
2862 {
2863         vmcs_write32(VM_ENTRY_CONTROLS,
2864                      vmcs_read32(VM_ENTRY_CONTROLS)
2865                      & ~VM_ENTRY_IA32E_MODE);
2866         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2867 }
2868
2869 #endif
2870
2871 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2872 {
2873         vpid_sync_context(to_vmx(vcpu));
2874         if (enable_ept) {
2875                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2876                         return;
2877                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2878         }
2879 }
2880
2881 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2882 {
2883         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2884
2885         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2886         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2887 }
2888
2889 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2890 {
2891         if (enable_ept && is_paging(vcpu))
2892                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2893         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2894 }
2895
2896 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2897 {
2898         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2899
2900         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2901         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2902 }
2903
2904 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2905 {
2906         if (!test_bit(VCPU_EXREG_PDPTR,
2907                       (unsigned long *)&vcpu->arch.regs_dirty))
2908                 return;
2909
2910         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2911                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2912                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2913                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2914                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2915         }
2916 }
2917
2918 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2919 {
2920         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2921                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2922                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2923                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2924                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2925         }
2926
2927         __set_bit(VCPU_EXREG_PDPTR,
2928                   (unsigned long *)&vcpu->arch.regs_avail);
2929         __set_bit(VCPU_EXREG_PDPTR,
2930                   (unsigned long *)&vcpu->arch.regs_dirty);
2931 }
2932
2933 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2934
2935 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2936                                         unsigned long cr0,
2937                                         struct kvm_vcpu *vcpu)
2938 {
2939         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2940                 vmx_decache_cr3(vcpu);
2941         if (!(cr0 & X86_CR0_PG)) {
2942                 /* From paging/starting to nonpaging */
2943                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2944                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2945                              (CPU_BASED_CR3_LOAD_EXITING |
2946                               CPU_BASED_CR3_STORE_EXITING));
2947                 vcpu->arch.cr0 = cr0;
2948                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2949         } else if (!is_paging(vcpu)) {
2950                 /* From nonpaging to paging */
2951                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2952                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2953                              ~(CPU_BASED_CR3_LOAD_EXITING |
2954                                CPU_BASED_CR3_STORE_EXITING));
2955                 vcpu->arch.cr0 = cr0;
2956                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2957         }
2958
2959         if (!(cr0 & X86_CR0_WP))
2960                 *hw_cr0 &= ~X86_CR0_WP;
2961 }
2962
2963 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2964 {
2965         struct vcpu_vmx *vmx = to_vmx(vcpu);
2966         unsigned long hw_cr0;
2967
2968         if (enable_unrestricted_guest)
2969                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2970                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2971         else
2972                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2973
2974         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2975                 enter_pmode(vcpu);
2976
2977         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2978                 enter_rmode(vcpu);
2979
2980 #ifdef CONFIG_X86_64
2981         if (vcpu->arch.efer & EFER_LME) {
2982                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2983                         enter_lmode(vcpu);
2984                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2985                         exit_lmode(vcpu);
2986         }
2987 #endif
2988
2989         if (enable_ept)
2990                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2991
2992         if (!vcpu->fpu_active)
2993                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2994
2995         vmcs_writel(CR0_READ_SHADOW, cr0);
2996         vmcs_writel(GUEST_CR0, hw_cr0);
2997         vcpu->arch.cr0 = cr0;
2998         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2999 }
3000
3001 static u64 construct_eptp(unsigned long root_hpa)
3002 {
3003         u64 eptp;
3004
3005         /* TODO write the value reading from MSR */
3006         eptp = VMX_EPT_DEFAULT_MT |
3007                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3008         eptp |= (root_hpa & PAGE_MASK);
3009
3010         return eptp;
3011 }
3012
3013 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3014 {
3015         unsigned long guest_cr3;
3016         u64 eptp;
3017
3018         guest_cr3 = cr3;
3019         if (enable_ept) {
3020                 eptp = construct_eptp(cr3);
3021                 vmcs_write64(EPT_POINTER, eptp);
3022                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3023                         vcpu->kvm->arch.ept_identity_map_addr;
3024                 ept_load_pdptrs(vcpu);
3025         }
3026
3027         vmx_flush_tlb(vcpu);
3028         vmcs_writel(GUEST_CR3, guest_cr3);
3029 }
3030
3031 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3032 {
3033         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3034                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3035
3036         if (cr4 & X86_CR4_VMXE) {
3037                 /*
3038                  * To use VMXON (and later other VMX instructions), a guest
3039                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3040                  * So basically the check on whether to allow nested VMX
3041                  * is here.
3042                  */
3043                 if (!nested_vmx_allowed(vcpu))
3044                         return 1;
3045         } else if (to_vmx(vcpu)->nested.vmxon)
3046                 return 1;
3047
3048         vcpu->arch.cr4 = cr4;
3049         if (enable_ept) {
3050                 if (!is_paging(vcpu)) {
3051                         hw_cr4 &= ~X86_CR4_PAE;
3052                         hw_cr4 |= X86_CR4_PSE;
3053                 } else if (!(cr4 & X86_CR4_PAE)) {
3054                         hw_cr4 &= ~X86_CR4_PAE;
3055                 }
3056         }
3057
3058         vmcs_writel(CR4_READ_SHADOW, cr4);
3059         vmcs_writel(GUEST_CR4, hw_cr4);
3060         return 0;
3061 }
3062
3063 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3064                             struct kvm_segment *var, int seg)
3065 {
3066         struct vcpu_vmx *vmx = to_vmx(vcpu);
3067         struct kvm_save_segment *save;
3068         u32 ar;
3069
3070         if (vmx->rmode.vm86_active
3071             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3072                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
3073                 || seg == VCPU_SREG_GS)
3074             && !emulate_invalid_guest_state) {
3075                 switch (seg) {
3076                 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
3077                 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
3078                 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
3079                 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
3080                 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
3081                 default: BUG();
3082                 }
3083                 var->selector = save->selector;
3084                 var->base = save->base;
3085                 var->limit = save->limit;
3086                 ar = save->ar;
3087                 if (seg == VCPU_SREG_TR
3088                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3089                         goto use_saved_rmode_seg;
3090         }
3091         var->base = vmx_read_guest_seg_base(vmx, seg);
3092         var->limit = vmx_read_guest_seg_limit(vmx, seg);
3093         var->selector = vmx_read_guest_seg_selector(vmx, seg);
3094         ar = vmx_read_guest_seg_ar(vmx, seg);
3095 use_saved_rmode_seg:
3096         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
3097                 ar = 0;
3098         var->type = ar & 15;
3099         var->s = (ar >> 4) & 1;
3100         var->dpl = (ar >> 5) & 3;
3101         var->present = (ar >> 7) & 1;
3102         var->avl = (ar >> 12) & 1;
3103         var->l = (ar >> 13) & 1;
3104         var->db = (ar >> 14) & 1;
3105         var->g = (ar >> 15) & 1;
3106         var->unusable = (ar >> 16) & 1;
3107 }
3108
3109 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3110 {
3111         struct kvm_segment s;
3112
3113         if (to_vmx(vcpu)->rmode.vm86_active) {
3114                 vmx_get_segment(vcpu, &s, seg);
3115                 return s.base;
3116         }
3117         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3118 }
3119
3120 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
3121 {
3122         if (!is_protmode(vcpu))
3123                 return 0;
3124
3125         if (!is_long_mode(vcpu)
3126             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3127                 return 3;
3128
3129         return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3130 }
3131
3132 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3133 {
3134         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3135                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3136                 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
3137         }
3138         return to_vmx(vcpu)->cpl;
3139 }
3140
3141
3142 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3143 {
3144         u32 ar;
3145
3146         if (var->unusable)
3147                 ar = 1 << 16;
3148         else {
3149                 ar = var->type & 15;
3150                 ar |= (var->s & 1) << 4;
3151                 ar |= (var->dpl & 3) << 5;
3152                 ar |= (var->present & 1) << 7;
3153                 ar |= (var->avl & 1) << 12;
3154                 ar |= (var->l & 1) << 13;
3155                 ar |= (var->db & 1) << 14;
3156                 ar |= (var->g & 1) << 15;
3157         }
3158         if (ar == 0) /* a 0 value means unusable */
3159                 ar = AR_UNUSABLE_MASK;
3160
3161         return ar;
3162 }
3163
3164 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3165                             struct kvm_segment *var, int seg)
3166 {
3167         struct vcpu_vmx *vmx = to_vmx(vcpu);
3168         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3169         u32 ar;
3170
3171         vmx_segment_cache_clear(vmx);
3172
3173         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3174                 vmcs_write16(sf->selector, var->selector);
3175                 vmx->rmode.tr.selector = var->selector;
3176                 vmx->rmode.tr.base = var->base;
3177                 vmx->rmode.tr.limit = var->limit;
3178                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
3179                 return;
3180         }
3181         vmcs_writel(sf->base, var->base);
3182         vmcs_write32(sf->limit, var->limit);
3183         vmcs_write16(sf->selector, var->selector);
3184         if (vmx->rmode.vm86_active && var->s) {
3185                 /*
3186                  * Hack real-mode segments into vm86 compatibility.
3187                  */
3188                 if (var->base == 0xffff0000 && var->selector == 0xf000)
3189                         vmcs_writel(sf->base, 0xf0000);
3190                 ar = 0xf3;
3191         } else
3192                 ar = vmx_segment_access_rights(var);
3193
3194         /*
3195          *   Fix the "Accessed" bit in AR field of segment registers for older
3196          * qemu binaries.
3197          *   IA32 arch specifies that at the time of processor reset the
3198          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3199          * is setting it to 0 in the usedland code. This causes invalid guest
3200          * state vmexit when "unrestricted guest" mode is turned on.
3201          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3202          * tree. Newer qemu binaries with that qemu fix would not need this
3203          * kvm hack.
3204          */
3205         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3206                 ar |= 0x1; /* Accessed */
3207
3208         vmcs_write32(sf->ar_bytes, ar);
3209         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3210 }
3211
3212 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3213 {
3214         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3215
3216         *db = (ar >> 14) & 1;
3217         *l = (ar >> 13) & 1;
3218 }
3219
3220 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3221 {
3222         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3223         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3224 }
3225
3226 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3227 {
3228         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3229         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3230 }
3231
3232 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3233 {
3234         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3235         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3236 }
3237
3238 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3239 {
3240         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3241         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3242 }
3243
3244 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3245 {
3246         struct kvm_segment var;
3247         u32 ar;
3248
3249         vmx_get_segment(vcpu, &var, seg);
3250         ar = vmx_segment_access_rights(&var);
3251
3252         if (var.base != (var.selector << 4))
3253                 return false;
3254         if (var.limit != 0xffff)
3255                 return false;
3256         if (ar != 0xf3)
3257                 return false;
3258
3259         return true;
3260 }
3261
3262 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3263 {
3264         struct kvm_segment cs;
3265         unsigned int cs_rpl;
3266
3267         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3268         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3269
3270         if (cs.unusable)
3271                 return false;
3272         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3273                 return false;
3274         if (!cs.s)
3275                 return false;
3276         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3277                 if (cs.dpl > cs_rpl)
3278                         return false;
3279         } else {
3280                 if (cs.dpl != cs_rpl)
3281                         return false;
3282         }
3283         if (!cs.present)
3284                 return false;
3285
3286         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3287         return true;
3288 }
3289
3290 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3291 {
3292         struct kvm_segment ss;
3293         unsigned int ss_rpl;
3294
3295         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3296         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3297
3298         if (ss.unusable)
3299                 return true;
3300         if (ss.type != 3 && ss.type != 7)
3301                 return false;
3302         if (!ss.s)
3303                 return false;
3304         if (ss.dpl != ss_rpl) /* DPL != RPL */
3305                 return false;
3306         if (!ss.present)
3307                 return false;
3308
3309         return true;
3310 }
3311
3312 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3313 {
3314         struct kvm_segment var;
3315         unsigned int rpl;
3316
3317         vmx_get_segment(vcpu, &var, seg);
3318         rpl = var.selector & SELECTOR_RPL_MASK;
3319
3320         if (var.unusable)
3321                 return true;
3322         if (!var.s)
3323                 return false;
3324         if (!var.present)
3325                 return false;
3326         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3327                 if (var.dpl < rpl) /* DPL < RPL */
3328                         return false;
3329         }
3330
3331         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3332          * rights flags
3333          */
3334         return true;
3335 }
3336
3337 static bool tr_valid(struct kvm_vcpu *vcpu)
3338 {
3339         struct kvm_segment tr;
3340
3341         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3342
3343         if (tr.unusable)
3344                 return false;
3345         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3346                 return false;
3347         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3348                 return false;
3349         if (!tr.present)
3350                 return false;
3351
3352         return true;
3353 }
3354
3355 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3356 {
3357         struct kvm_segment ldtr;
3358
3359         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3360
3361         if (ldtr.unusable)
3362                 return true;
3363         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3364                 return false;
3365         if (ldtr.type != 2)
3366                 return false;
3367         if (!ldtr.present)
3368                 return false;
3369
3370         return true;
3371 }
3372
3373 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3374 {
3375         struct kvm_segment cs, ss;
3376
3377         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3378         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3379
3380         return ((cs.selector & SELECTOR_RPL_MASK) ==
3381                  (ss.selector & SELECTOR_RPL_MASK));
3382 }
3383
3384 /*
3385  * Check if guest state is valid. Returns true if valid, false if
3386  * not.
3387  * We assume that registers are always usable
3388  */
3389 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3390 {
3391         /* real mode guest state checks */
3392         if (!is_protmode(vcpu)) {
3393                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3394                         return false;
3395                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3396                         return false;
3397                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3398                         return false;
3399                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3400                         return false;
3401                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3402                         return false;
3403                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3404                         return false;
3405         } else {
3406         /* protected mode guest state checks */
3407                 if (!cs_ss_rpl_check(vcpu))
3408                         return false;
3409                 if (!code_segment_valid(vcpu))
3410                         return false;
3411                 if (!stack_segment_valid(vcpu))
3412                         return false;
3413                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3414                         return false;
3415                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3416                         return false;
3417                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3418                         return false;
3419                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3420                         return false;
3421                 if (!tr_valid(vcpu))
3422                         return false;
3423                 if (!ldtr_valid(vcpu))
3424                         return false;
3425         }
3426         /* TODO:
3427          * - Add checks on RIP
3428          * - Add checks on RFLAGS
3429          */
3430
3431         return true;
3432 }
3433
3434 static int init_rmode_tss(struct kvm *kvm)
3435 {
3436         gfn_t fn;
3437         u16 data = 0;
3438         int r, idx, ret = 0;
3439
3440         idx = srcu_read_lock(&kvm->srcu);
3441         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3442         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3443         if (r < 0)
3444                 goto out;
3445         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3446         r = kvm_write_guest_page(kvm, fn++, &data,
3447                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3448         if (r < 0)
3449                 goto out;
3450         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3451         if (r < 0)
3452                 goto out;
3453         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3454         if (r < 0)
3455                 goto out;
3456         data = ~0;
3457         r = kvm_write_guest_page(kvm, fn, &data,
3458                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3459                                  sizeof(u8));
3460         if (r < 0)
3461                 goto out;
3462
3463         ret = 1;
3464 out:
3465         srcu_read_unlock(&kvm->srcu, idx);
3466         return ret;
3467 }
3468
3469 static int init_rmode_identity_map(struct kvm *kvm)
3470 {
3471         int i, idx, r, ret;
3472         pfn_t identity_map_pfn;
3473         u32 tmp;
3474
3475         if (!enable_ept)
3476                 return 1;
3477         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3478                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3479                         "haven't been allocated!\n");
3480                 return 0;
3481         }
3482         if (likely(kvm->arch.ept_identity_pagetable_done))
3483                 return 1;
3484         ret = 0;
3485         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3486         idx = srcu_read_lock(&kvm->srcu);
3487         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3488         if (r < 0)
3489                 goto out;
3490         /* Set up identity-mapping pagetable for EPT in real mode */
3491         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3492                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3493                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3494                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3495                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3496                 if (r < 0)
3497                         goto out;
3498         }
3499         kvm->arch.ept_identity_pagetable_done = true;
3500         ret = 1;
3501 out:
3502         srcu_read_unlock(&kvm->srcu, idx);
3503         return ret;
3504 }
3505
3506 static void seg_setup(int seg)
3507 {
3508         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3509         unsigned int ar;
3510
3511         vmcs_write16(sf->selector, 0);
3512         vmcs_writel(sf->base, 0);
3513         vmcs_write32(sf->limit, 0xffff);
3514         if (enable_unrestricted_guest) {
3515                 ar = 0x93;
3516                 if (seg == VCPU_SREG_CS)
3517                         ar |= 0x08; /* code segment */
3518         } else
3519                 ar = 0xf3;
3520
3521         vmcs_write32(sf->ar_bytes, ar);
3522 }
3523
3524 static int alloc_apic_access_page(struct kvm *kvm)
3525 {
3526         struct kvm_userspace_memory_region kvm_userspace_mem;
3527         int r = 0;
3528
3529         mutex_lock(&kvm->slots_lock);
3530         if (kvm->arch.apic_access_page)
3531                 goto out;
3532         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3533         kvm_userspace_mem.flags = 0;
3534         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3535         kvm_userspace_mem.memory_size = PAGE_SIZE;
3536         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3537         if (r)
3538                 goto out;
3539
3540         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3541 out:
3542         mutex_unlock(&kvm->slots_lock);
3543         return r;
3544 }
3545
3546 static int alloc_identity_pagetable(struct kvm *kvm)
3547 {
3548         struct kvm_userspace_memory_region kvm_userspace_mem;
3549         int r = 0;
3550
3551         mutex_lock(&kvm->slots_lock);
3552         if (kvm->arch.ept_identity_pagetable)
3553                 goto out;
3554         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3555         kvm_userspace_mem.flags = 0;
3556         kvm_userspace_mem.guest_phys_addr =
3557                 kvm->arch.ept_identity_map_addr;
3558         kvm_userspace_mem.memory_size = PAGE_SIZE;
3559         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3560         if (r)
3561                 goto out;
3562
3563         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3564                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3565 out:
3566         mutex_unlock(&kvm->slots_lock);
3567         return r;
3568 }
3569
3570 static void allocate_vpid(struct vcpu_vmx *vmx)
3571 {
3572         int vpid;
3573
3574         vmx->vpid = 0;
3575         if (!enable_vpid)
3576                 return;
3577         spin_lock(&vmx_vpid_lock);
3578         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3579         if (vpid < VMX_NR_VPIDS) {
3580                 vmx->vpid = vpid;
3581                 __set_bit(vpid, vmx_vpid_bitmap);
3582         }
3583         spin_unlock(&vmx_vpid_lock);
3584 }
3585
3586 static void free_vpid(struct vcpu_vmx *vmx)
3587 {
3588         if (!enable_vpid)
3589                 return;
3590         spin_lock(&vmx_vpid_lock);
3591         if (vmx->vpid != 0)
3592                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3593         spin_unlock(&vmx_vpid_lock);
3594 }
3595
3596 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3597 {
3598         int f = sizeof(unsigned long);
3599
3600         if (!cpu_has_vmx_msr_bitmap())
3601                 return;
3602
3603         /*
3604          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3605          * have the write-low and read-high bitmap offsets the wrong way round.
3606          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3607          */
3608         if (msr <= 0x1fff) {
3609                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3610                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3611         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3612                 msr &= 0x1fff;
3613                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3614                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3615         }
3616 }
3617
3618 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3619 {
3620         if (!longmode_only)
3621                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3622         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3623 }
3624
3625 /*
3626  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3627  * will not change in the lifetime of the guest.
3628  * Note that host-state that does change is set elsewhere. E.g., host-state
3629  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3630  */
3631 static void vmx_set_constant_host_state(void)
3632 {
3633         u32 low32, high32;
3634         unsigned long tmpl;
3635         struct desc_ptr dt;
3636
3637         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
3638         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
3639         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
3640
3641         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3642         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3643         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3644         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3645         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3646
3647         native_store_idt(&dt);
3648         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3649
3650         asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3651         vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3652
3653         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3654         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3655         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3656         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3657
3658         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3659                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3660                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3661         }
3662 }
3663
3664 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3665 {
3666         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3667         if (enable_ept)
3668                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3669         if (is_guest_mode(&vmx->vcpu))
3670                 vmx->vcpu.arch.cr4_guest_owned_bits &=
3671                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3672         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3673 }
3674
3675 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3676 {
3677         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3678         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3679                 exec_control &= ~CPU_BASED_TPR_SHADOW;
3680 #ifdef CONFIG_X86_64
3681                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3682                                 CPU_BASED_CR8_LOAD_EXITING;
3683 #endif
3684         }
3685         if (!enable_ept)
3686                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3687                                 CPU_BASED_CR3_LOAD_EXITING  |
3688                                 CPU_BASED_INVLPG_EXITING;
3689         return exec_control;
3690 }
3691
3692 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3693 {
3694         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3695         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3696                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3697         if (vmx->vpid == 0)
3698                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3699         if (!enable_ept) {
3700                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3701                 enable_unrestricted_guest = 0;
3702         }
3703         if (!enable_unrestricted_guest)
3704                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3705         if (!ple_gap)
3706                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3707         return exec_control;
3708 }
3709
3710 static void ept_set_mmio_spte_mask(void)
3711 {
3712         /*
3713          * EPT Misconfigurations can be generated if the value of bits 2:0
3714          * of an EPT paging-structure entry is 110b (write/execute).
3715          * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3716          * spte.
3717          */
3718         kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3719 }
3720
3721 /*
3722  * Sets up the vmcs for emulated real mode.
3723  */
3724 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3725 {
3726 #ifdef CONFIG_X86_64
3727         unsigned long a;
3728 #endif
3729         int i;
3730
3731         /* I/O */
3732         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3733         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3734
3735         if (cpu_has_vmx_msr_bitmap())
3736                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3737
3738         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3739
3740         /* Control */
3741         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3742                 vmcs_config.pin_based_exec_ctrl);
3743
3744         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3745
3746         if (cpu_has_secondary_exec_ctrls()) {
3747                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3748                                 vmx_secondary_exec_control(vmx));
3749         }
3750
3751         if (ple_gap) {
3752                 vmcs_write32(PLE_GAP, ple_gap);
3753                 vmcs_write32(PLE_WINDOW, ple_window);
3754         }
3755
3756         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3757         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
3758         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
3759
3760         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
3761         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
3762         vmx_set_constant_host_state();
3763 #ifdef CONFIG_X86_64
3764         rdmsrl(MSR_FS_BASE, a);
3765         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3766         rdmsrl(MSR_GS_BASE, a);
3767         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3768 #else
3769         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3770         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3771 #endif
3772
3773         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3774         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3775         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3776         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3777         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3778
3779         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3780                 u32 msr_low, msr_high;
3781                 u64 host_pat;
3782                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3783                 host_pat = msr_low | ((u64) msr_high << 32);
3784                 /* Write the default value follow host pat */
3785                 vmcs_write64(GUEST_IA32_PAT, host_pat);
3786                 /* Keep arch.pat sync with GUEST_IA32_PAT */
3787                 vmx->vcpu.arch.pat = host_pat;
3788         }
3789
3790         for (i = 0; i < NR_VMX_MSR; ++i) {
3791                 u32 index = vmx_msr_index[i];
3792                 u32 data_low, data_high;
3793                 int j = vmx->nmsrs;
3794
3795                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3796                         continue;
3797                 if (wrmsr_safe(index, data_low, data_high) < 0)
3798                         continue;
3799                 vmx->guest_msrs[j].index = i;
3800                 vmx->guest_msrs[j].data = 0;
3801                 vmx->guest_msrs[j].mask = -1ull;
3802                 ++vmx->nmsrs;
3803         }
3804
3805         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3806
3807         /* 22.2.1, 20.8.1 */
3808         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3809
3810         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3811         set_cr4_guest_host_mask(vmx);
3812
3813         kvm_write_tsc(&vmx->vcpu, 0);
3814
3815         return 0;
3816 }
3817
3818 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3819 {
3820         struct vcpu_vmx *vmx = to_vmx(vcpu);
3821         u64 msr;
3822         int ret;
3823
3824         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3825
3826         vmx->rmode.vm86_active = 0;
3827
3828         vmx->soft_vnmi_blocked = 0;
3829
3830         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3831         kvm_set_cr8(&vmx->vcpu, 0);
3832         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3833         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3834                 msr |= MSR_IA32_APICBASE_BSP;
3835         kvm_set_apic_base(&vmx->vcpu, msr);
3836
3837         ret = fx_init(&vmx->vcpu);
3838         if (ret != 0)
3839                 goto out;
3840
3841         vmx_segment_cache_clear(vmx);
3842
3843         seg_setup(VCPU_SREG_CS);
3844         /*
3845          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3846          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
3847          */
3848         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3849                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3850                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3851         } else {
3852                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3853                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3854         }
3855
3856         seg_setup(VCPU_SREG_DS);
3857         seg_setup(VCPU_SREG_ES);
3858         seg_setup(VCPU_SREG_FS);
3859         seg_setup(VCPU_SREG_GS);
3860         seg_setup(VCPU_SREG_SS);
3861
3862         vmcs_write16(GUEST_TR_SELECTOR, 0);
3863         vmcs_writel(GUEST_TR_BASE, 0);
3864         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3865         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3866
3867         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3868         vmcs_writel(GUEST_LDTR_BASE, 0);
3869         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3870         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3871
3872         vmcs_write32(GUEST_SYSENTER_CS, 0);
3873         vmcs_writel(GUEST_SYSENTER_ESP, 0);
3874         vmcs_writel(GUEST_SYSENTER_EIP, 0);
3875
3876         vmcs_writel(GUEST_RFLAGS, 0x02);
3877         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3878                 kvm_rip_write(vcpu, 0xfff0);
3879         else
3880                 kvm_rip_write(vcpu, 0);
3881         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3882
3883         vmcs_writel(GUEST_DR7, 0x400);
3884
3885         vmcs_writel(GUEST_GDTR_BASE, 0);
3886         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3887
3888         vmcs_writel(GUEST_IDTR_BASE, 0);
3889         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3890
3891         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3892         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3893         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3894
3895         /* Special registers */
3896         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3897
3898         setup_msrs(vmx);
3899
3900         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
3901
3902         if (cpu_has_vmx_tpr_shadow()) {
3903                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3904                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3905                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3906                                      __pa(vmx->vcpu.arch.apic->regs));
3907                 vmcs_write32(TPR_THRESHOLD, 0);
3908         }
3909
3910         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3911                 vmcs_write64(APIC_ACCESS_ADDR,
3912                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3913
3914         if (vmx->vpid != 0)
3915                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3916
3917         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3918         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3919         vmx_set_cr4(&vmx->vcpu, 0);
3920         vmx_set_efer(&vmx->vcpu, 0);
3921         vmx_fpu_activate(&vmx->vcpu);
3922         update_exception_bitmap(&vmx->vcpu);
3923
3924         vpid_sync_context(vmx);
3925
3926         ret = 0;
3927
3928         /* HACK: Don't enable emulation on guest boot/reset */
3929         vmx->emulation_required = 0;
3930
3931 out:
3932         return ret;
3933 }
3934
3935 /*
3936  * In nested virtualization, check if L1 asked to exit on external interrupts.
3937  * For most existing hypervisors, this will always return true.
3938  */
3939 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
3940 {
3941         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
3942                 PIN_BASED_EXT_INTR_MASK;
3943 }
3944
3945 static void enable_irq_window(struct kvm_vcpu *vcpu)
3946 {
3947         u32 cpu_based_vm_exec_control;
3948         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
3949                 /*
3950                  * We get here if vmx_interrupt_allowed() said we can't
3951                  * inject to L1 now because L2 must run. Ask L2 to exit
3952                  * right after entry, so we can inject to L1 more promptly.
3953                  */
3954                 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
3955                 return;
3956         }
3957
3958         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3959         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3960         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3961 }
3962
3963 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3964 {
3965         u32 cpu_based_vm_exec_control;
3966
3967         if (!cpu_has_virtual_nmis()) {
3968                 enable_irq_window(vcpu);
3969                 return;
3970         }
3971
3972         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3973                 enable_irq_window(vcpu);
3974                 return;
3975         }
3976         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3977         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3978         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3979 }
3980
3981 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3982 {
3983         struct vcpu_vmx *vmx = to_vmx(vcpu);
3984         uint32_t intr;
3985         int irq = vcpu->arch.interrupt.nr;
3986
3987         trace_kvm_inj_virq(irq);
3988
3989         ++vcpu->stat.irq_injections;
3990         if (vmx->rmode.vm86_active) {
3991                 int inc_eip = 0;
3992                 if (vcpu->arch.interrupt.soft)
3993                         inc_eip = vcpu->arch.event_exit_inst_len;
3994                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3995                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3996                 return;
3997         }
3998         intr = irq | INTR_INFO_VALID_MASK;
3999         if (vcpu->arch.interrupt.soft) {
4000                 intr |= INTR_TYPE_SOFT_INTR;
4001                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4002                              vmx->vcpu.arch.event_exit_inst_len);
4003         } else
4004                 intr |= INTR_TYPE_EXT_INTR;
4005         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4006         vmx_clear_hlt(vcpu);
4007 }
4008
4009 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4010 {
4011         struct vcpu_vmx *vmx = to_vmx(vcpu);
4012
4013         if (is_guest_mode(vcpu))
4014                 return;
4015
4016         if (!cpu_has_virtual_nmis()) {
4017                 /*
4018                  * Tracking the NMI-blocked state in software is built upon
4019                  * finding the next open IRQ window. This, in turn, depends on
4020                  * well-behaving guests: They have to keep IRQs disabled at
4021                  * least as long as the NMI handler runs. Otherwise we may
4022                  * cause NMI nesting, maybe breaking the guest. But as this is
4023                  * highly unlikely, we can live with the residual risk.
4024                  */
4025                 vmx->soft_vnmi_blocked = 1;
4026                 vmx->vnmi_blocked_time = 0;
4027         }
4028
4029         ++vcpu->stat.nmi_injections;
4030         vmx->nmi_known_unmasked = false;
4031         if (vmx->rmode.vm86_active) {
4032                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4033                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4034                 return;
4035         }
4036         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4037                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4038         vmx_clear_hlt(vcpu);
4039 }
4040
4041 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4042 {
4043         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4044                 return 0;
4045
4046         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4047                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4048                    | GUEST_INTR_STATE_NMI));
4049 }
4050
4051 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4052 {
4053         if (!cpu_has_virtual_nmis())
4054                 return to_vmx(vcpu)->soft_vnmi_blocked;
4055         if (to_vmx(vcpu)->nmi_known_unmasked)
4056                 return false;
4057         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4058 }
4059
4060 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4061 {
4062         struct vcpu_vmx *vmx = to_vmx(vcpu);
4063
4064         if (!cpu_has_virtual_nmis()) {
4065                 if (vmx->soft_vnmi_blocked != masked) {
4066                         vmx->soft_vnmi_blocked = masked;
4067                         vmx->vnmi_blocked_time = 0;
4068                 }
4069         } else {
4070                 vmx->nmi_known_unmasked = !masked;
4071                 if (masked)
4072                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4073                                       GUEST_INTR_STATE_NMI);
4074                 else
4075                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4076                                         GUEST_INTR_STATE_NMI);
4077         }
4078 }
4079
4080 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4081 {
4082         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4083                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4084                 if (to_vmx(vcpu)->nested.nested_run_pending ||
4085                     (vmcs12->idt_vectoring_info_field &
4086                      VECTORING_INFO_VALID_MASK))
4087                         return 0;
4088                 nested_vmx_vmexit(vcpu);
4089                 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4090                 vmcs12->vm_exit_intr_info = 0;
4091                 /* fall through to normal code, but now in L1, not L2 */
4092         }
4093
4094         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4095                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4096                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4097 }
4098
4099 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4100 {
4101         int ret;
4102         struct kvm_userspace_memory_region tss_mem = {
4103                 .slot = TSS_PRIVATE_MEMSLOT,
4104                 .guest_phys_addr = addr,
4105                 .memory_size = PAGE_SIZE * 3,
4106                 .flags = 0,
4107         };
4108
4109         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4110         if (ret)
4111                 return ret;
4112         kvm->arch.tss_addr = addr;
4113         if (!init_rmode_tss(kvm))
4114                 return  -ENOMEM;
4115
4116         return 0;
4117 }
4118
4119 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4120                                   int vec, u32 err_code)
4121 {
4122         /*
4123          * Instruction with address size override prefix opcode 0x67
4124          * Cause the #SS fault with 0 error code in VM86 mode.
4125          */
4126         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
4127                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
4128                         return 1;
4129         /*
4130          * Forward all other exceptions that are valid in real mode.
4131          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4132          *        the required debugging infrastructure rework.
4133          */
4134         switch (vec) {
4135         case DB_VECTOR:
4136                 if (vcpu->guest_debug &
4137                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4138                         return 0;
4139                 kvm_queue_exception(vcpu, vec);
4140                 return 1;
4141         case BP_VECTOR:
4142                 /*
4143                  * Update instruction length as we may reinject the exception
4144                  * from user space while in guest debugging mode.
4145                  */
4146                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4147                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4148                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4149                         return 0;
4150                 /* fall through */
4151         case DE_VECTOR:
4152         case OF_VECTOR:
4153         case BR_VECTOR:
4154         case UD_VECTOR:
4155         case DF_VECTOR:
4156         case SS_VECTOR:
4157         case GP_VECTOR:
4158         case MF_VECTOR:
4159                 kvm_queue_exception(vcpu, vec);
4160                 return 1;
4161         }
4162         return 0;
4163 }
4164
4165 /*
4166  * Trigger machine check on the host. We assume all the MSRs are already set up
4167  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4168  * We pass a fake environment to the machine check handler because we want
4169  * the guest to be always treated like user space, no matter what context
4170  * it used internally.
4171  */
4172 static void kvm_machine_check(void)
4173 {
4174 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4175         struct pt_regs regs = {
4176                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4177                 .flags = X86_EFLAGS_IF,
4178         };
4179
4180         do_machine_check(&regs, 0);
4181 #endif
4182 }
4183
4184 static int handle_machine_check(struct kvm_vcpu *vcpu)
4185 {
4186         /* already handled by vcpu_run */
4187         return 1;
4188 }
4189
4190 static int handle_exception(struct kvm_vcpu *vcpu)
4191 {
4192         struct vcpu_vmx *vmx = to_vmx(vcpu);
4193         struct kvm_run *kvm_run = vcpu->run;
4194         u32 intr_info, ex_no, error_code;
4195         unsigned long cr2, rip, dr6;
4196         u32 vect_info;
4197         enum emulation_result er;
4198
4199         vect_info = vmx->idt_vectoring_info;
4200         intr_info = vmx->exit_intr_info;
4201
4202         if (is_machine_check(intr_info))
4203                 return handle_machine_check(vcpu);
4204
4205         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4206             !is_page_fault(intr_info)) {
4207                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4208                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4209                 vcpu->run->internal.ndata = 2;
4210                 vcpu->run->internal.data[0] = vect_info;
4211                 vcpu->run->internal.data[1] = intr_info;
4212                 return 0;
4213         }
4214
4215         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4216                 return 1;  /* already handled by vmx_vcpu_run() */
4217
4218         if (is_no_device(intr_info)) {
4219                 vmx_fpu_activate(vcpu);
4220                 return 1;
4221         }
4222
4223         if (is_invalid_opcode(intr_info)) {
4224                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4225                 if (er != EMULATE_DONE)
4226                         kvm_queue_exception(vcpu, UD_VECTOR);
4227                 return 1;
4228         }
4229
4230         error_code = 0;
4231         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4232                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4233         if (is_page_fault(intr_info)) {
4234                 /* EPT won't cause page fault directly */
4235                 BUG_ON(enable_ept);
4236                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4237                 trace_kvm_page_fault(cr2, error_code);
4238
4239                 if (kvm_event_needs_reinjection(vcpu))
4240                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4241                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4242         }
4243
4244         if (vmx->rmode.vm86_active &&
4245             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4246                                                                 error_code)) {
4247                 if (vcpu->arch.halt_request) {
4248                         vcpu->arch.halt_request = 0;
4249                         return kvm_emulate_halt(vcpu);
4250                 }
4251                 return 1;
4252         }
4253
4254         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4255         switch (ex_no) {
4256         case DB_VECTOR:
4257                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4258                 if (!(vcpu->guest_debug &
4259                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4260                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4261                         kvm_queue_exception(vcpu, DB_VECTOR);
4262                         return 1;
4263                 }
4264                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4265                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4266                 /* fall through */
4267         case BP_VECTOR:
4268                 /*
4269                  * Update instruction length as we may reinject #BP from
4270                  * user space while in guest debugging mode. Reading it for
4271                  * #DB as well causes no harm, it is not used in that case.
4272                  */
4273                 vmx->vcpu.arch.event_exit_inst_len =
4274                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4275                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4276                 rip = kvm_rip_read(vcpu);
4277                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4278                 kvm_run->debug.arch.exception = ex_no;
4279                 break;
4280         default:
4281                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4282                 kvm_run->ex.exception = ex_no;
4283                 kvm_run->ex.error_code = error_code;
4284                 break;
4285         }
4286         return 0;
4287 }
4288
4289 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4290 {
4291         ++vcpu->stat.irq_exits;
4292         return 1;
4293 }
4294
4295 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4296 {
4297         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4298         return 0;
4299 }
4300
4301 static int handle_io(struct kvm_vcpu *vcpu)
4302 {
4303         unsigned long exit_qualification;
4304         int size, in, string;
4305         unsigned port;
4306
4307         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4308         string = (exit_qualification & 16) != 0;
4309         in = (exit_qualification & 8) != 0;
4310
4311         ++vcpu->stat.io_exits;
4312
4313         if (string || in)
4314                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4315
4316         port = exit_qualification >> 16;
4317         size = (exit_qualification & 7) + 1;
4318         skip_emulated_instruction(vcpu);
4319
4320         return kvm_fast_pio_out(vcpu, size, port);
4321 }
4322
4323 static void
4324 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4325 {
4326         /*
4327          * Patch in the VMCALL instruction:
4328          */
4329         hypercall[0] = 0x0f;
4330         hypercall[1] = 0x01;
4331         hypercall[2] = 0xc1;
4332 }
4333
4334 /* called to set cr0 as approriate for a mov-to-cr0 exit. */
4335 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4336 {
4337         if (to_vmx(vcpu)->nested.vmxon &&
4338             ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4339                 return 1;
4340
4341         if (is_guest_mode(vcpu)) {
4342                 /*
4343                  * We get here when L2 changed cr0 in a way that did not change
4344                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4345                  * but did change L0 shadowed bits. This can currently happen
4346                  * with the TS bit: L0 may want to leave TS on (for lazy fpu
4347                  * loading) while pretending to allow the guest to change it.
4348                  */
4349                 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4350                          (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4351                         return 1;
4352                 vmcs_writel(CR0_READ_SHADOW, val);
4353                 return 0;
4354         } else
4355                 return kvm_set_cr0(vcpu, val);
4356 }
4357
4358 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4359 {
4360         if (is_guest_mode(vcpu)) {
4361                 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4362                          (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4363                         return 1;
4364                 vmcs_writel(CR4_READ_SHADOW, val);
4365                 return 0;
4366         } else
4367                 return kvm_set_cr4(vcpu, val);
4368 }
4369
4370 /* called to set cr0 as approriate for clts instruction exit. */
4371 static void handle_clts(struct kvm_vcpu *vcpu)
4372 {
4373         if (is_guest_mode(vcpu)) {
4374                 /*
4375                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4376                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4377                  * just pretend it's off (also in arch.cr0 for fpu_activate).
4378                  */
4379                 vmcs_writel(CR0_READ_SHADOW,
4380                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4381                 vcpu->arch.cr0 &= ~X86_CR0_TS;
4382         } else
4383                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4384 }
4385
4386 static int handle_cr(struct kvm_vcpu *vcpu)
4387 {
4388         unsigned long exit_qualification, val;
4389         int cr;
4390         int reg;
4391         int err;
4392
4393         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4394         cr = exit_qualification & 15;
4395         reg = (exit_qualification >> 8) & 15;
4396         switch ((exit_qualification >> 4) & 3) {
4397         case 0: /* mov to cr */
4398                 val = kvm_register_read(vcpu, reg);
4399                 trace_kvm_cr_write(cr, val);
4400                 switch (cr) {
4401                 case 0:
4402                         err = handle_set_cr0(vcpu, val);
4403                         kvm_complete_insn_gp(vcpu, err);
4404                         return 1;
4405                 case 3:
4406                         err = kvm_set_cr3(vcpu, val);
4407                         kvm_complete_insn_gp(vcpu, err);
4408                         return 1;
4409                 case 4:
4410                         err = handle_set_cr4(vcpu, val);
4411                         kvm_complete_insn_gp(vcpu, err);
4412                         return 1;
4413                 case 8: {
4414                                 u8 cr8_prev = kvm_get_cr8(vcpu);
4415                                 u8 cr8 = kvm_register_read(vcpu, reg);
4416                                 err = kvm_set_cr8(vcpu, cr8);
4417                                 kvm_complete_insn_gp(vcpu, err);
4418                                 if (irqchip_in_kernel(vcpu->kvm))
4419                                         return 1;
4420                                 if (cr8_prev <= cr8)
4421                                         return 1;
4422                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4423                                 return 0;
4424                         }
4425                 };
4426                 break;
4427         case 2: /* clts */
4428                 handle_clts(vcpu);
4429                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4430                 skip_emulated_instruction(vcpu);
4431                 vmx_fpu_activate(vcpu);
4432                 return 1;
4433         case 1: /*mov from cr*/
4434                 switch (cr) {
4435                 case 3:
4436                         val = kvm_read_cr3(vcpu);
4437                         kvm_register_write(vcpu, reg, val);
4438                         trace_kvm_cr_read(cr, val);
4439                         skip_emulated_instruction(vcpu);
4440                         return 1;
4441                 case 8:
4442                         val = kvm_get_cr8(vcpu);
4443                         kvm_register_write(vcpu, reg, val);
4444                         trace_kvm_cr_read(cr, val);
4445                         skip_emulated_instruction(vcpu);
4446                         return 1;
4447                 }
4448                 break;
4449         case 3: /* lmsw */
4450                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4451                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4452                 kvm_lmsw(vcpu, val);
4453
4454                 skip_emulated_instruction(vcpu);
4455                 return 1;
4456         default:
4457                 break;
4458         }
4459         vcpu->run->exit_reason = 0;
4460         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4461                (int)(exit_qualification >> 4) & 3, cr);
4462         return 0;
4463 }
4464
4465 static int handle_dr(struct kvm_vcpu *vcpu)
4466 {
4467         unsigned long exit_qualification;
4468         int dr, reg;
4469
4470         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4471         if (!kvm_require_cpl(vcpu, 0))
4472                 return 1;
4473         dr = vmcs_readl(GUEST_DR7);
4474         if (dr & DR7_GD) {
4475                 /*
4476                  * As the vm-exit takes precedence over the debug trap, we
4477                  * need to emulate the latter, either for the host or the
4478                  * guest debugging itself.
4479                  */
4480                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4481                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4482                         vcpu->run->debug.arch.dr7 = dr;
4483                         vcpu->run->debug.arch.pc =
4484                                 vmcs_readl(GUEST_CS_BASE) +
4485                                 vmcs_readl(GUEST_RIP);
4486                         vcpu->run->debug.arch.exception = DB_VECTOR;
4487                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4488                         return 0;
4489                 } else {
4490                         vcpu->arch.dr7 &= ~DR7_GD;
4491                         vcpu->arch.dr6 |= DR6_BD;
4492                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4493                         kvm_queue_exception(vcpu, DB_VECTOR);
4494                         return 1;
4495                 }
4496         }
4497
4498         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4499         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4500         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4501         if (exit_qualification & TYPE_MOV_FROM_DR) {
4502                 unsigned long val;
4503                 if (!kvm_get_dr(vcpu, dr, &val))
4504                         kvm_register_write(vcpu, reg, val);
4505         } else
4506                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4507         skip_emulated_instruction(vcpu);
4508         return 1;
4509 }
4510
4511 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4512 {
4513         vmcs_writel(GUEST_DR7, val);
4514 }
4515
4516 static int handle_cpuid(struct kvm_vcpu *vcpu)
4517 {
4518         kvm_emulate_cpuid(vcpu);
4519         return 1;
4520 }
4521
4522 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4523 {
4524         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4525         u64 data;
4526
4527         if (vmx_get_msr(vcpu, ecx, &data)) {
4528                 trace_kvm_msr_read_ex(ecx);
4529                 kvm_inject_gp(vcpu, 0);
4530                 return 1;
4531         }
4532
4533         trace_kvm_msr_read(ecx, data);
4534
4535         /* FIXME: handling of bits 32:63 of rax, rdx */
4536         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4537         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4538         skip_emulated_instruction(vcpu);
4539         return 1;
4540 }
4541
4542 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4543 {
4544         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4545         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4546                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4547
4548         if (vmx_set_msr(vcpu, ecx, data) != 0) {
4549                 trace_kvm_msr_write_ex(ecx, data);
4550                 kvm_inject_gp(vcpu, 0);
4551                 return 1;
4552         }
4553
4554         trace_kvm_msr_write(ecx, data);
4555         skip_emulated_instruction(vcpu);
4556         return 1;
4557 }
4558
4559 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4560 {
4561         kvm_make_request(KVM_REQ_EVENT, vcpu);
4562         return 1;
4563 }
4564
4565 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4566 {
4567         u32 cpu_based_vm_exec_control;
4568
4569         /* clear pending irq */
4570         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4571         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4572         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4573
4574         kvm_make_request(KVM_REQ_EVENT, vcpu);
4575
4576         ++vcpu->stat.irq_window_exits;
4577
4578         /*
4579          * If the user space waits to inject interrupts, exit as soon as
4580          * possible
4581          */
4582         if (!irqchip_in_kernel(vcpu->kvm) &&
4583             vcpu->run->request_interrupt_window &&
4584             !kvm_cpu_has_interrupt(vcpu)) {
4585                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4586                 return 0;
4587         }
4588         return 1;
4589 }
4590
4591 static int handle_halt(struct kvm_vcpu *vcpu)
4592 {
4593         skip_emulated_instruction(vcpu);
4594         return kvm_emulate_halt(vcpu);
4595 }
4596
4597 static int handle_vmcall(struct kvm_vcpu *vcpu)
4598 {
4599         skip_emulated_instruction(vcpu);
4600         kvm_emulate_hypercall(vcpu);
4601         return 1;
4602 }
4603
4604 static int handle_invd(struct kvm_vcpu *vcpu)
4605 {
4606         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4607 }
4608
4609 static int handle_invlpg(struct kvm_vcpu *vcpu)
4610 {
4611         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4612
4613         kvm_mmu_invlpg(vcpu, exit_qualification);
4614         skip_emulated_instruction(vcpu);
4615         return 1;
4616 }
4617
4618 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4619 {
4620         int err;
4621
4622         err = kvm_rdpmc(vcpu);
4623         kvm_complete_insn_gp(vcpu, err);
4624
4625         return 1;
4626 }
4627
4628 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4629 {
4630         skip_emulated_instruction(vcpu);
4631         kvm_emulate_wbinvd(vcpu);
4632         return 1;
4633 }
4634
4635 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4636 {
4637         u64 new_bv = kvm_read_edx_eax(vcpu);
4638         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4639
4640         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4641                 skip_emulated_instruction(vcpu);
4642         return 1;
4643 }
4644
4645 static int handle_apic_access(struct kvm_vcpu *vcpu)
4646 {
4647         if (likely(fasteoi)) {
4648                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4649                 int access_type, offset;
4650
4651                 access_type = exit_qualification & APIC_ACCESS_TYPE;
4652                 offset = exit_qualification & APIC_ACCESS_OFFSET;
4653                 /*
4654                  * Sane guest uses MOV to write EOI, with written value
4655                  * not cared. So make a short-circuit here by avoiding
4656                  * heavy instruction emulation.
4657                  */
4658                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4659                     (offset == APIC_EOI)) {
4660                         kvm_lapic_set_eoi(vcpu);
4661                         skip_emulated_instruction(vcpu);
4662                         return 1;
4663                 }
4664         }
4665         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4666 }
4667
4668 static int handle_task_switch(struct kvm_vcpu *vcpu)
4669 {
4670         struct vcpu_vmx *vmx = to_vmx(vcpu);
4671         unsigned long exit_qualification;
4672         bool has_error_code = false;
4673         u32 error_code = 0;
4674         u16 tss_selector;
4675         int reason, type, idt_v;
4676
4677         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4678         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4679
4680         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4681
4682         reason = (u32)exit_qualification >> 30;
4683         if (reason == TASK_SWITCH_GATE && idt_v) {
4684                 switch (type) {
4685                 case INTR_TYPE_NMI_INTR:
4686                         vcpu->arch.nmi_injected = false;
4687                         vmx_set_nmi_mask(vcpu, true);
4688                         break;
4689                 case INTR_TYPE_EXT_INTR:
4690                 case INTR_TYPE_SOFT_INTR:
4691                         kvm_clear_interrupt_queue(vcpu);
4692                         break;
4693                 case INTR_TYPE_HARD_EXCEPTION:
4694                         if (vmx->idt_vectoring_info &
4695                             VECTORING_INFO_DELIVER_CODE_MASK) {
4696                                 has_error_code = true;
4697                                 error_code =
4698                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
4699                         }
4700                         /* fall through */
4701                 case INTR_TYPE_SOFT_EXCEPTION:
4702                         kvm_clear_exception_queue(vcpu);
4703                         break;
4704                 default:
4705                         break;
4706                 }
4707         }
4708         tss_selector = exit_qualification;
4709
4710         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4711                        type != INTR_TYPE_EXT_INTR &&
4712                        type != INTR_TYPE_NMI_INTR))
4713                 skip_emulated_instruction(vcpu);
4714
4715         if (kvm_task_switch(vcpu, tss_selector, reason,
4716                                 has_error_code, error_code) == EMULATE_FAIL) {
4717                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4718                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4719                 vcpu->run->internal.ndata = 0;
4720                 return 0;
4721         }
4722
4723         /* clear all local breakpoint enable flags */
4724         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4725
4726         /*
4727          * TODO: What about debug traps on tss switch?
4728          *       Are we supposed to inject them and update dr6?
4729          */
4730
4731         return 1;
4732 }
4733
4734 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4735 {
4736         unsigned long exit_qualification;
4737         gpa_t gpa;
4738         int gla_validity;
4739
4740         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4741
4742         if (exit_qualification & (1 << 6)) {
4743                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4744                 return -EINVAL;
4745         }
4746
4747         gla_validity = (exit_qualification >> 7) & 0x3;
4748         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4749                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4750                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4751                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4752                         vmcs_readl(GUEST_LINEAR_ADDRESS));
4753                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4754                         (long unsigned int)exit_qualification);
4755                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4756                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4757                 return 0;
4758         }
4759
4760         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4761         trace_kvm_page_fault(gpa, exit_qualification);
4762         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4763 }
4764
4765 static u64 ept_rsvd_mask(u64 spte, int level)
4766 {
4767         int i;
4768         u64 mask = 0;
4769
4770         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4771                 mask |= (1ULL << i);
4772
4773         if (level > 2)
4774                 /* bits 7:3 reserved */
4775                 mask |= 0xf8;
4776         else if (level == 2) {
4777                 if (spte & (1ULL << 7))
4778                         /* 2MB ref, bits 20:12 reserved */
4779                         mask |= 0x1ff000;
4780                 else
4781                         /* bits 6:3 reserved */
4782                         mask |= 0x78;
4783         }
4784
4785         return mask;
4786 }
4787
4788 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4789                                        int level)
4790 {
4791         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4792
4793         /* 010b (write-only) */
4794         WARN_ON((spte & 0x7) == 0x2);
4795
4796         /* 110b (write/execute) */
4797         WARN_ON((spte & 0x7) == 0x6);
4798
4799         /* 100b (execute-only) and value not supported by logical processor */
4800         if (!cpu_has_vmx_ept_execute_only())
4801                 WARN_ON((spte & 0x7) == 0x4);
4802
4803         /* not 000b */
4804         if ((spte & 0x7)) {
4805                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4806
4807                 if (rsvd_bits != 0) {
4808                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4809                                          __func__, rsvd_bits);
4810                         WARN_ON(1);
4811                 }
4812
4813                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4814                         u64 ept_mem_type = (spte & 0x38) >> 3;
4815
4816                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
4817                             ept_mem_type == 7) {
4818                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4819                                                 __func__, ept_mem_type);
4820                                 WARN_ON(1);
4821                         }
4822                 }
4823         }
4824 }
4825
4826 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4827 {
4828         u64 sptes[4];
4829         int nr_sptes, i, ret;
4830         gpa_t gpa;
4831
4832         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4833
4834         ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4835         if (likely(ret == 1))
4836                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4837                                               EMULATE_DONE;
4838         if (unlikely(!ret))
4839                 return 1;
4840
4841         /* It is the real ept misconfig */
4842         printk(KERN_ERR "EPT: Misconfiguration.\n");
4843         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4844
4845         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4846
4847         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4848                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4849
4850         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4851         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4852
4853         return 0;
4854 }
4855
4856 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4857 {
4858         u32 cpu_based_vm_exec_control;
4859
4860         /* clear pending NMI */
4861         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4862         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4863         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4864         ++vcpu->stat.nmi_window_exits;
4865         kvm_make_request(KVM_REQ_EVENT, vcpu);
4866
4867         return 1;
4868 }
4869
4870 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4871 {
4872         struct vcpu_vmx *vmx = to_vmx(vcpu);
4873         enum emulation_result err = EMULATE_DONE;
4874         int ret = 1;
4875         u32 cpu_exec_ctrl;
4876         bool intr_window_requested;
4877
4878         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4879         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4880
4881         while (!guest_state_valid(vcpu)) {
4882                 if (intr_window_requested
4883                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4884                         return handle_interrupt_window(&vmx->vcpu);
4885
4886                 err = emulate_instruction(vcpu, 0);
4887
4888                 if (err == EMULATE_DO_MMIO) {
4889                         ret = 0;
4890                         goto out;
4891                 }
4892
4893                 if (err != EMULATE_DONE)
4894                         return 0;
4895
4896                 if (signal_pending(current))
4897                         goto out;
4898                 if (need_resched())
4899                         schedule();
4900         }
4901
4902         vmx->emulation_required = 0;
4903 out:
4904         return ret;
4905 }
4906
4907 /*
4908  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4909  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4910  */
4911 static int handle_pause(struct kvm_vcpu *vcpu)
4912 {
4913         skip_emulated_instruction(vcpu);
4914         kvm_vcpu_on_spin(vcpu);
4915
4916         return 1;
4917 }
4918
4919 static int handle_invalid_op(struct kvm_vcpu *vcpu)
4920 {
4921         kvm_queue_exception(vcpu, UD_VECTOR);
4922         return 1;
4923 }
4924
4925 /*
4926  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
4927  * We could reuse a single VMCS for all the L2 guests, but we also want the
4928  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
4929  * allows keeping them loaded on the processor, and in the future will allow
4930  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
4931  * every entry if they never change.
4932  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
4933  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
4934  *
4935  * The following functions allocate and free a vmcs02 in this pool.
4936  */
4937
4938 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
4939 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
4940 {
4941         struct vmcs02_list *item;
4942         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4943                 if (item->vmptr == vmx->nested.current_vmptr) {
4944                         list_move(&item->list, &vmx->nested.vmcs02_pool);
4945                         return &item->vmcs02;
4946                 }
4947
4948         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
4949                 /* Recycle the least recently used VMCS. */
4950                 item = list_entry(vmx->nested.vmcs02_pool.prev,
4951                         struct vmcs02_list, list);
4952                 item->vmptr = vmx->nested.current_vmptr;
4953                 list_move(&item->list, &vmx->nested.vmcs02_pool);
4954                 return &item->vmcs02;
4955         }
4956
4957         /* Create a new VMCS */
4958         item = (struct vmcs02_list *)
4959                 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
4960         if (!item)
4961                 return NULL;
4962         item->vmcs02.vmcs = alloc_vmcs();
4963         if (!item->vmcs02.vmcs) {
4964                 kfree(item);
4965                 return NULL;
4966         }
4967         loaded_vmcs_init(&item->vmcs02);
4968         item->vmptr = vmx->nested.current_vmptr;
4969         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
4970         vmx->nested.vmcs02_num++;
4971         return &item->vmcs02;
4972 }
4973
4974 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
4975 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
4976 {
4977         struct vmcs02_list *item;
4978         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4979                 if (item->vmptr == vmptr) {
4980                         free_loaded_vmcs(&item->vmcs02);
4981                         list_del(&item->list);
4982                         kfree(item);
4983                         vmx->nested.vmcs02_num--;
4984                         return;
4985                 }
4986 }
4987
4988 /*
4989  * Free all VMCSs saved for this vcpu, except the one pointed by
4990  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
4991  * currently used, if running L2), and vmcs01 when running L2.
4992  */
4993 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
4994 {
4995         struct vmcs02_list *item, *n;
4996         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
4997                 if (vmx->loaded_vmcs != &item->vmcs02)
4998                         free_loaded_vmcs(&item->vmcs02);
4999                 list_del(&item->list);
5000                 kfree(item);
5001         }
5002         vmx->nested.vmcs02_num = 0;
5003
5004         if (vmx->loaded_vmcs != &vmx->vmcs01)
5005                 free_loaded_vmcs(&vmx->vmcs01);
5006 }
5007
5008 /*
5009  * Emulate the VMXON instruction.
5010  * Currently, we just remember that VMX is active, and do not save or even
5011  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5012  * do not currently need to store anything in that guest-allocated memory
5013  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5014  * argument is different from the VMXON pointer (which the spec says they do).
5015  */
5016 static int handle_vmon(struct kvm_vcpu *vcpu)
5017 {
5018         struct kvm_segment cs;
5019         struct vcpu_vmx *vmx = to_vmx(vcpu);
5020
5021         /* The Intel VMX Instruction Reference lists a bunch of bits that
5022          * are prerequisite to running VMXON, most notably cr4.VMXE must be
5023          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5024          * Otherwise, we should fail with #UD. We test these now:
5025          */
5026         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5027             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5028             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5029                 kvm_queue_exception(vcpu, UD_VECTOR);
5030                 return 1;
5031         }
5032
5033         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5034         if (is_long_mode(vcpu) && !cs.l) {
5035                 kvm_queue_exception(vcpu, UD_VECTOR);
5036                 return 1;
5037         }
5038
5039         if (vmx_get_cpl(vcpu)) {
5040                 kvm_inject_gp(vcpu, 0);
5041                 return 1;
5042         }
5043
5044         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5045         vmx->nested.vmcs02_num = 0;
5046
5047         vmx->nested.vmxon = true;
5048
5049         skip_emulated_instruction(vcpu);
5050         return 1;
5051 }
5052
5053 /*
5054  * Intel's VMX Instruction Reference specifies a common set of prerequisites
5055  * for running VMX instructions (except VMXON, whose prerequisites are
5056  * slightly different). It also specifies what exception to inject otherwise.
5057  */
5058 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5059 {
5060         struct kvm_segment cs;
5061         struct vcpu_vmx *vmx = to_vmx(vcpu);
5062
5063         if (!vmx->nested.vmxon) {
5064                 kvm_queue_exception(vcpu, UD_VECTOR);
5065                 return 0;
5066         }
5067
5068         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5069         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5070             (is_long_mode(vcpu) && !cs.l)) {
5071                 kvm_queue_exception(vcpu, UD_VECTOR);
5072                 return 0;
5073         }
5074
5075         if (vmx_get_cpl(vcpu)) {
5076                 kvm_inject_gp(vcpu, 0);
5077                 return 0;
5078         }
5079
5080         return 1;
5081 }
5082
5083 /*
5084  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5085  * just stops using VMX.
5086  */
5087 static void free_nested(struct vcpu_vmx *vmx)
5088 {
5089         if (!vmx->nested.vmxon)
5090                 return;
5091         vmx->nested.vmxon = false;
5092         if (vmx->nested.current_vmptr != -1ull) {
5093                 kunmap(vmx->nested.current_vmcs12_page);
5094                 nested_release_page(vmx->nested.current_vmcs12_page);
5095                 vmx->nested.current_vmptr = -1ull;
5096                 vmx->nested.current_vmcs12 = NULL;
5097         }
5098         /* Unpin physical memory we referred to in current vmcs02 */
5099         if (vmx->nested.apic_access_page) {
5100                 nested_release_page(vmx->nested.apic_access_page);
5101                 vmx->nested.apic_access_page = 0;
5102         }
5103
5104         nested_free_all_saved_vmcss(vmx);
5105 }
5106
5107 /* Emulate the VMXOFF instruction */
5108 static int handle_vmoff(struct kvm_vcpu *vcpu)
5109 {
5110         if (!nested_vmx_check_permission(vcpu))
5111                 return 1;
5112         free_nested(to_vmx(vcpu));
5113         skip_emulated_instruction(vcpu);
5114         return 1;
5115 }
5116
5117 /*
5118  * Decode the memory-address operand of a vmx instruction, as recorded on an
5119  * exit caused by such an instruction (run by a guest hypervisor).
5120  * On success, returns 0. When the operand is invalid, returns 1 and throws
5121  * #UD or #GP.
5122  */
5123 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5124                                  unsigned long exit_qualification,
5125                                  u32 vmx_instruction_info, gva_t *ret)
5126 {
5127         /*
5128          * According to Vol. 3B, "Information for VM Exits Due to Instruction
5129          * Execution", on an exit, vmx_instruction_info holds most of the
5130          * addressing components of the operand. Only the displacement part
5131          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5132          * For how an actual address is calculated from all these components,
5133          * refer to Vol. 1, "Operand Addressing".
5134          */
5135         int  scaling = vmx_instruction_info & 3;
5136         int  addr_size = (vmx_instruction_info >> 7) & 7;
5137         bool is_reg = vmx_instruction_info & (1u << 10);
5138         int  seg_reg = (vmx_instruction_info >> 15) & 7;
5139         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5140         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5141         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5142         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5143
5144         if (is_reg) {
5145                 kvm_queue_exception(vcpu, UD_VECTOR);
5146                 return 1;
5147         }
5148
5149         /* Addr = segment_base + offset */
5150         /* offset = base + [index * scale] + displacement */
5151         *ret = vmx_get_segment_base(vcpu, seg_reg);
5152         if (base_is_valid)
5153                 *ret += kvm_register_read(vcpu, base_reg);
5154         if (index_is_valid)
5155                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5156         *ret += exit_qualification; /* holds the displacement */
5157
5158         if (addr_size == 1) /* 32 bit */
5159                 *ret &= 0xffffffff;
5160
5161         /*
5162          * TODO: throw #GP (and return 1) in various cases that the VM*
5163          * instructions require it - e.g., offset beyond segment limit,
5164          * unusable or unreadable/unwritable segment, non-canonical 64-bit
5165          * address, and so on. Currently these are not checked.
5166          */
5167         return 0;
5168 }
5169
5170 /*
5171  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5172  * set the success or error code of an emulated VMX instruction, as specified
5173  * by Vol 2B, VMX Instruction Reference, "Conventions".
5174  */
5175 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5176 {
5177         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5178                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5179                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5180 }
5181
5182 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5183 {
5184         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5185                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5186                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5187                         | X86_EFLAGS_CF);
5188 }
5189
5190 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5191                                         u32 vm_instruction_error)
5192 {
5193         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5194                 /*
5195                  * failValid writes the error number to the current VMCS, which
5196                  * can't be done there isn't a current VMCS.
5197                  */
5198                 nested_vmx_failInvalid(vcpu);
5199                 return;
5200         }
5201         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5202                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5203                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5204                         | X86_EFLAGS_ZF);
5205         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5206 }
5207
5208 /* Emulate the VMCLEAR instruction */
5209 static int handle_vmclear(struct kvm_vcpu *vcpu)
5210 {
5211         struct vcpu_vmx *vmx = to_vmx(vcpu);
5212         gva_t gva;
5213         gpa_t vmptr;
5214         struct vmcs12 *vmcs12;
5215         struct page *page;
5216         struct x86_exception e;
5217
5218         if (!nested_vmx_check_permission(vcpu))
5219                 return 1;
5220
5221         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5222                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5223                 return 1;
5224
5225         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5226                                 sizeof(vmptr), &e)) {
5227                 kvm_inject_page_fault(vcpu, &e);
5228                 return 1;
5229         }
5230
5231         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5232                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5233                 skip_emulated_instruction(vcpu);
5234                 return 1;
5235         }
5236
5237         if (vmptr == vmx->nested.current_vmptr) {
5238                 kunmap(vmx->nested.current_vmcs12_page);
5239                 nested_release_page(vmx->nested.current_vmcs12_page);
5240                 vmx->nested.current_vmptr = -1ull;
5241                 vmx->nested.current_vmcs12 = NULL;
5242         }
5243
5244         page = nested_get_page(vcpu, vmptr);
5245         if (page == NULL) {
5246                 /*
5247                  * For accurate processor emulation, VMCLEAR beyond available
5248                  * physical memory should do nothing at all. However, it is
5249                  * possible that a nested vmx bug, not a guest hypervisor bug,
5250                  * resulted in this case, so let's shut down before doing any
5251                  * more damage:
5252                  */
5253                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5254                 return 1;
5255         }
5256         vmcs12 = kmap(page);
5257         vmcs12->launch_state = 0;
5258         kunmap(page);
5259         nested_release_page(page);
5260
5261         nested_free_vmcs02(vmx, vmptr);
5262
5263         skip_emulated_instruction(vcpu);
5264         nested_vmx_succeed(vcpu);
5265         return 1;
5266 }
5267
5268 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5269
5270 /* Emulate the VMLAUNCH instruction */
5271 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5272 {
5273         return nested_vmx_run(vcpu, true);
5274 }
5275
5276 /* Emulate the VMRESUME instruction */
5277 static int handle_vmresume(struct kvm_vcpu *vcpu)
5278 {
5279
5280         return nested_vmx_run(vcpu, false);
5281 }
5282
5283 enum vmcs_field_type {
5284         VMCS_FIELD_TYPE_U16 = 0,
5285         VMCS_FIELD_TYPE_U64 = 1,
5286         VMCS_FIELD_TYPE_U32 = 2,
5287         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5288 };
5289
5290 static inline int vmcs_field_type(unsigned long field)
5291 {
5292         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
5293                 return VMCS_FIELD_TYPE_U32;
5294         return (field >> 13) & 0x3 ;
5295 }
5296
5297 static inline int vmcs_field_readonly(unsigned long field)
5298 {
5299         return (((field >> 10) & 0x3) == 1);
5300 }
5301
5302 /*
5303  * Read a vmcs12 field. Since these can have varying lengths and we return
5304  * one type, we chose the biggest type (u64) and zero-extend the return value
5305  * to that size. Note that the caller, handle_vmread, might need to use only
5306  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5307  * 64-bit fields are to be returned).
5308  */
5309 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5310                                         unsigned long field, u64 *ret)
5311 {
5312         short offset = vmcs_field_to_offset(field);
5313         char *p;
5314
5315         if (offset < 0)
5316                 return 0;
5317
5318         p = ((char *)(get_vmcs12(vcpu))) + offset;
5319
5320         switch (vmcs_field_type(field)) {
5321         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5322                 *ret = *((natural_width *)p);
5323                 return 1;
5324         case VMCS_FIELD_TYPE_U16:
5325                 *ret = *((u16 *)p);
5326                 return 1;
5327         case VMCS_FIELD_TYPE_U32:
5328                 *ret = *((u32 *)p);
5329                 return 1;
5330         case VMCS_FIELD_TYPE_U64:
5331                 *ret = *((u64 *)p);
5332                 return 1;
5333         default:
5334                 return 0; /* can never happen. */
5335         }
5336 }
5337
5338 /*
5339  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5340  * used before) all generate the same failure when it is missing.
5341  */
5342 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5343 {
5344         struct vcpu_vmx *vmx = to_vmx(vcpu);
5345         if (vmx->nested.current_vmptr == -1ull) {
5346                 nested_vmx_failInvalid(vcpu);
5347                 skip_emulated_instruction(vcpu);
5348                 return 0;
5349         }
5350         return 1;
5351 }
5352
5353 static int handle_vmread(struct kvm_vcpu *vcpu)
5354 {
5355         unsigned long field;
5356         u64 field_value;
5357         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5358         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5359         gva_t gva = 0;
5360
5361         if (!nested_vmx_check_permission(vcpu) ||
5362             !nested_vmx_check_vmcs12(vcpu))
5363                 return 1;
5364
5365         /* Decode instruction info and find the field to read */
5366         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5367         /* Read the field, zero-extended to a u64 field_value */
5368         if (!vmcs12_read_any(vcpu, field, &field_value)) {
5369                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5370                 skip_emulated_instruction(vcpu);
5371                 return 1;
5372         }
5373         /*
5374          * Now copy part of this value to register or memory, as requested.
5375          * Note that the number of bits actually copied is 32 or 64 depending
5376          * on the guest's mode (32 or 64 bit), not on the given field's length.
5377          */
5378         if (vmx_instruction_info & (1u << 10)) {
5379                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5380                         field_value);
5381         } else {
5382                 if (get_vmx_mem_address(vcpu, exit_qualification,
5383                                 vmx_instruction_info, &gva))
5384                         return 1;
5385                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5386                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5387                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5388         }
5389
5390         nested_vmx_succeed(vcpu);
5391         skip_emulated_instruction(vcpu);
5392         return 1;
5393 }
5394
5395
5396 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5397 {
5398         unsigned long field;
5399         gva_t gva;
5400         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5401         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5402         char *p;
5403         short offset;
5404         /* The value to write might be 32 or 64 bits, depending on L1's long
5405          * mode, and eventually we need to write that into a field of several
5406          * possible lengths. The code below first zero-extends the value to 64
5407          * bit (field_value), and then copies only the approriate number of
5408          * bits into the vmcs12 field.
5409          */
5410         u64 field_value = 0;
5411         struct x86_exception e;
5412
5413         if (!nested_vmx_check_permission(vcpu) ||
5414             !nested_vmx_check_vmcs12(vcpu))
5415                 return 1;
5416
5417         if (vmx_instruction_info & (1u << 10))
5418                 field_value = kvm_register_read(vcpu,
5419                         (((vmx_instruction_info) >> 3) & 0xf));
5420         else {
5421                 if (get_vmx_mem_address(vcpu, exit_qualification,
5422                                 vmx_instruction_info, &gva))
5423                         return 1;
5424                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5425                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5426                         kvm_inject_page_fault(vcpu, &e);
5427                         return 1;
5428                 }
5429         }
5430
5431
5432         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5433         if (vmcs_field_readonly(field)) {
5434                 nested_vmx_failValid(vcpu,
5435                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5436                 skip_emulated_instruction(vcpu);
5437                 return 1;
5438         }
5439
5440         offset = vmcs_field_to_offset(field);
5441         if (offset < 0) {
5442                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5443                 skip_emulated_instruction(vcpu);
5444                 return 1;
5445         }
5446         p = ((char *) get_vmcs12(vcpu)) + offset;
5447
5448         switch (vmcs_field_type(field)) {
5449         case VMCS_FIELD_TYPE_U16:
5450                 *(u16 *)p = field_value;
5451                 break;
5452         case VMCS_FIELD_TYPE_U32:
5453                 *(u32 *)p = field_value;
5454                 break;
5455         case VMCS_FIELD_TYPE_U64:
5456                 *(u64 *)p = field_value;
5457                 break;
5458         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5459                 *(natural_width *)p = field_value;
5460                 break;
5461         default:
5462                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5463                 skip_emulated_instruction(vcpu);
5464                 return 1;
5465         }
5466
5467         nested_vmx_succeed(vcpu);
5468         skip_emulated_instruction(vcpu);
5469         return 1;
5470 }
5471
5472 /* Emulate the VMPTRLD instruction */
5473 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5474 {
5475         struct vcpu_vmx *vmx = to_vmx(vcpu);
5476         gva_t gva;
5477         gpa_t vmptr;
5478         struct x86_exception e;
5479
5480         if (!nested_vmx_check_permission(vcpu))
5481                 return 1;
5482
5483         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5484                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5485                 return 1;
5486
5487         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5488                                 sizeof(vmptr), &e)) {
5489                 kvm_inject_page_fault(vcpu, &e);
5490                 return 1;
5491         }
5492
5493         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5494                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5495                 skip_emulated_instruction(vcpu);
5496                 return 1;
5497         }
5498
5499         if (vmx->nested.current_vmptr != vmptr) {
5500                 struct vmcs12 *new_vmcs12;
5501                 struct page *page;
5502                 page = nested_get_page(vcpu, vmptr);
5503                 if (page == NULL) {
5504                         nested_vmx_failInvalid(vcpu);
5505                         skip_emulated_instruction(vcpu);
5506                         return 1;
5507                 }
5508                 new_vmcs12 = kmap(page);
5509                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5510                         kunmap(page);
5511                         nested_release_page_clean(page);
5512                         nested_vmx_failValid(vcpu,
5513                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5514                         skip_emulated_instruction(vcpu);
5515                         return 1;
5516                 }
5517                 if (vmx->nested.current_vmptr != -1ull) {
5518                         kunmap(vmx->nested.current_vmcs12_page);
5519                         nested_release_page(vmx->nested.current_vmcs12_page);
5520                 }
5521
5522                 vmx->nested.current_vmptr = vmptr;
5523                 vmx->nested.current_vmcs12 = new_vmcs12;
5524                 vmx->nested.current_vmcs12_page = page;
5525         }
5526
5527         nested_vmx_succeed(vcpu);
5528         skip_emulated_instruction(vcpu);
5529         return 1;
5530 }
5531
5532 /* Emulate the VMPTRST instruction */
5533 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5534 {
5535         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5536         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5537         gva_t vmcs_gva;
5538         struct x86_exception e;
5539
5540         if (!nested_vmx_check_permission(vcpu))
5541                 return 1;
5542
5543         if (get_vmx_mem_address(vcpu, exit_qualification,
5544                         vmx_instruction_info, &vmcs_gva))
5545                 return 1;
5546         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5547         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5548                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
5549                                  sizeof(u64), &e)) {
5550                 kvm_inject_page_fault(vcpu, &e);
5551                 return 1;
5552         }
5553         nested_vmx_succeed(vcpu);
5554         skip_emulated_instruction(vcpu);
5555         return 1;
5556 }
5557
5558 /*
5559  * The exit handlers return 1 if the exit was handled fully and guest execution
5560  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5561  * to be done to userspace and return 0.
5562  */
5563 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5564         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5565         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5566         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5567         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
5568         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5569         [EXIT_REASON_CR_ACCESS]               = handle_cr,
5570         [EXIT_REASON_DR_ACCESS]               = handle_dr,
5571         [EXIT_REASON_CPUID]                   = handle_cpuid,
5572         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
5573         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5574         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5575         [EXIT_REASON_HLT]                     = handle_halt,
5576         [EXIT_REASON_INVD]                    = handle_invd,
5577         [EXIT_REASON_INVLPG]                  = handle_invlpg,
5578         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
5579         [EXIT_REASON_VMCALL]                  = handle_vmcall,
5580         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
5581         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5582         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5583         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5584         [EXIT_REASON_VMREAD]                  = handle_vmread,
5585         [EXIT_REASON_VMRESUME]                = handle_vmresume,
5586         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5587         [EXIT_REASON_VMOFF]                   = handle_vmoff,
5588         [EXIT_REASON_VMON]                    = handle_vmon,
5589         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5590         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5591         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
5592         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
5593         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5594         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5595         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
5596         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5597         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5598         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
5599         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5600 };
5601
5602 static const int kvm_vmx_max_exit_handlers =
5603         ARRAY_SIZE(kvm_vmx_exit_handlers);
5604
5605 /*
5606  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5607  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5608  * disinterest in the current event (read or write a specific MSR) by using an
5609  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5610  */
5611 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5612         struct vmcs12 *vmcs12, u32 exit_reason)
5613 {
5614         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5615         gpa_t bitmap;
5616
5617         if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5618                 return 1;
5619
5620         /*
5621          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5622          * for the four combinations of read/write and low/high MSR numbers.
5623          * First we need to figure out which of the four to use:
5624          */
5625         bitmap = vmcs12->msr_bitmap;
5626         if (exit_reason == EXIT_REASON_MSR_WRITE)
5627                 bitmap += 2048;
5628         if (msr_index >= 0xc0000000) {
5629                 msr_index -= 0xc0000000;
5630                 bitmap += 1024;
5631         }
5632
5633         /* Then read the msr_index'th bit from this bitmap: */
5634         if (msr_index < 1024*8) {
5635                 unsigned char b;
5636                 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5637                 return 1 & (b >> (msr_index & 7));
5638         } else
5639                 return 1; /* let L1 handle the wrong parameter */
5640 }
5641
5642 /*
5643  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5644  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5645  * intercept (via guest_host_mask etc.) the current event.
5646  */
5647 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5648         struct vmcs12 *vmcs12)
5649 {
5650         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5651         int cr = exit_qualification & 15;
5652         int reg = (exit_qualification >> 8) & 15;
5653         unsigned long val = kvm_register_read(vcpu, reg);
5654
5655         switch ((exit_qualification >> 4) & 3) {
5656         case 0: /* mov to cr */
5657                 switch (cr) {
5658                 case 0:
5659                         if (vmcs12->cr0_guest_host_mask &
5660                             (val ^ vmcs12->cr0_read_shadow))
5661                                 return 1;
5662                         break;
5663                 case 3:
5664                         if ((vmcs12->cr3_target_count >= 1 &&
5665                                         vmcs12->cr3_target_value0 == val) ||
5666                                 (vmcs12->cr3_target_count >= 2 &&
5667                                         vmcs12->cr3_target_value1 == val) ||
5668                                 (vmcs12->cr3_target_count >= 3 &&
5669                                         vmcs12->cr3_target_value2 == val) ||
5670                                 (vmcs12->cr3_target_count >= 4 &&
5671                                         vmcs12->cr3_target_value3 == val))
5672                                 return 0;
5673                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5674                                 return 1;
5675                         break;
5676                 case 4:
5677                         if (vmcs12->cr4_guest_host_mask &
5678                             (vmcs12->cr4_read_shadow ^ val))
5679                                 return 1;
5680                         break;
5681                 case 8:
5682                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5683                                 return 1;
5684                         break;
5685                 }
5686                 break;
5687         case 2: /* clts */
5688                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5689                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
5690                         return 1;
5691                 break;
5692         case 1: /* mov from cr */
5693                 switch (cr) {
5694                 case 3:
5695                         if (vmcs12->cpu_based_vm_exec_control &
5696                             CPU_BASED_CR3_STORE_EXITING)
5697                                 return 1;
5698                         break;
5699                 case 8:
5700                         if (vmcs12->cpu_based_vm_exec_control &
5701                             CPU_BASED_CR8_STORE_EXITING)
5702                                 return 1;
5703                         break;
5704                 }
5705                 break;
5706         case 3: /* lmsw */
5707                 /*
5708                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5709                  * cr0. Other attempted changes are ignored, with no exit.
5710                  */
5711                 if (vmcs12->cr0_guest_host_mask & 0xe &
5712                     (val ^ vmcs12->cr0_read_shadow))
5713                         return 1;
5714                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5715                     !(vmcs12->cr0_read_shadow & 0x1) &&
5716                     (val & 0x1))
5717                         return 1;
5718                 break;
5719         }
5720         return 0;
5721 }
5722
5723 /*
5724  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5725  * should handle it ourselves in L0 (and then continue L2). Only call this
5726  * when in is_guest_mode (L2).
5727  */
5728 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5729 {
5730         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5731         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5732         struct vcpu_vmx *vmx = to_vmx(vcpu);
5733         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5734
5735         if (vmx->nested.nested_run_pending)
5736                 return 0;
5737
5738         if (unlikely(vmx->fail)) {
5739                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5740                                     vmcs_read32(VM_INSTRUCTION_ERROR));
5741                 return 1;
5742         }
5743
5744         switch (exit_reason) {
5745         case EXIT_REASON_EXCEPTION_NMI:
5746                 if (!is_exception(intr_info))
5747                         return 0;
5748                 else if (is_page_fault(intr_info))
5749                         return enable_ept;
5750                 return vmcs12->exception_bitmap &
5751                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5752         case EXIT_REASON_EXTERNAL_INTERRUPT:
5753                 return 0;
5754         case EXIT_REASON_TRIPLE_FAULT:
5755                 return 1;
5756         case EXIT_REASON_PENDING_INTERRUPT:
5757         case EXIT_REASON_NMI_WINDOW:
5758                 /*
5759                  * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5760                  * (aka Interrupt Window Exiting) only when L1 turned it on,
5761                  * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5762                  * Same for NMI Window Exiting.
5763                  */
5764                 return 1;
5765         case EXIT_REASON_TASK_SWITCH:
5766                 return 1;
5767         case EXIT_REASON_CPUID:
5768                 return 1;
5769         case EXIT_REASON_HLT:
5770                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5771         case EXIT_REASON_INVD:
5772                 return 1;
5773         case EXIT_REASON_INVLPG:
5774                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5775         case EXIT_REASON_RDPMC:
5776                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5777         case EXIT_REASON_RDTSC:
5778                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5779         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5780         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5781         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5782         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5783         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5784                 /*
5785                  * VMX instructions trap unconditionally. This allows L1 to
5786                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
5787                  */
5788                 return 1;
5789         case EXIT_REASON_CR_ACCESS:
5790                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5791         case EXIT_REASON_DR_ACCESS:
5792                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5793         case EXIT_REASON_IO_INSTRUCTION:
5794                 /* TODO: support IO bitmaps */
5795                 return 1;
5796         case EXIT_REASON_MSR_READ:
5797         case EXIT_REASON_MSR_WRITE:
5798                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5799         case EXIT_REASON_INVALID_STATE:
5800                 return 1;
5801         case EXIT_REASON_MWAIT_INSTRUCTION:
5802                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5803         case EXIT_REASON_MONITOR_INSTRUCTION:
5804                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5805         case EXIT_REASON_PAUSE_INSTRUCTION:
5806                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5807                         nested_cpu_has2(vmcs12,
5808                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5809         case EXIT_REASON_MCE_DURING_VMENTRY:
5810                 return 0;
5811         case EXIT_REASON_TPR_BELOW_THRESHOLD:
5812                 return 1;
5813         case EXIT_REASON_APIC_ACCESS:
5814                 return nested_cpu_has2(vmcs12,
5815                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5816         case EXIT_REASON_EPT_VIOLATION:
5817         case EXIT_REASON_EPT_MISCONFIG:
5818                 return 0;
5819         case EXIT_REASON_WBINVD:
5820                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5821         case EXIT_REASON_XSETBV:
5822                 return 1;
5823         default:
5824                 return 1;
5825         }
5826 }
5827
5828 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5829 {
5830         *info1 = vmcs_readl(EXIT_QUALIFICATION);
5831         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5832 }
5833
5834 /*
5835  * The guest has exited.  See if we can fix it or if we need userspace
5836  * assistance.
5837  */
5838 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5839 {
5840         struct vcpu_vmx *vmx = to_vmx(vcpu);
5841         u32 exit_reason = vmx->exit_reason;
5842         u32 vectoring_info = vmx->idt_vectoring_info;
5843
5844         /* If guest state is invalid, start emulating */
5845         if (vmx->emulation_required && emulate_invalid_guest_state)
5846                 return handle_invalid_guest_state(vcpu);
5847
5848         /*
5849          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5850          * we did not inject a still-pending event to L1 now because of
5851          * nested_run_pending, we need to re-enable this bit.
5852          */
5853         if (vmx->nested.nested_run_pending)
5854                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5855
5856         if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5857             exit_reason == EXIT_REASON_VMRESUME))
5858                 vmx->nested.nested_run_pending = 1;
5859         else
5860                 vmx->nested.nested_run_pending = 0;
5861
5862         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5863                 nested_vmx_vmexit(vcpu);
5864                 return 1;
5865         }
5866
5867         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5868                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5869                 vcpu->run->fail_entry.hardware_entry_failure_reason
5870                         = exit_reason;
5871                 return 0;
5872         }
5873
5874         if (unlikely(vmx->fail)) {
5875                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5876                 vcpu->run->fail_entry.hardware_entry_failure_reason
5877                         = vmcs_read32(VM_INSTRUCTION_ERROR);
5878                 return 0;
5879         }
5880
5881         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5882                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5883                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
5884                         exit_reason != EXIT_REASON_TASK_SWITCH))
5885                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5886                        "(0x%x) and exit reason is 0x%x\n",
5887                        __func__, vectoring_info, exit_reason);
5888
5889         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5890             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5891                                         get_vmcs12(vcpu), vcpu)))) {
5892                 if (vmx_interrupt_allowed(vcpu)) {
5893                         vmx->soft_vnmi_blocked = 0;
5894                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5895                            vcpu->arch.nmi_pending) {
5896                         /*
5897                          * This CPU don't support us in finding the end of an
5898                          * NMI-blocked window if the guest runs with IRQs
5899                          * disabled. So we pull the trigger after 1 s of
5900                          * futile waiting, but inform the user about this.
5901                          */
5902                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5903                                "state on VCPU %d after 1 s timeout\n",
5904                                __func__, vcpu->vcpu_id);
5905                         vmx->soft_vnmi_blocked = 0;
5906                 }
5907         }
5908
5909         if (exit_reason < kvm_vmx_max_exit_handlers
5910             && kvm_vmx_exit_handlers[exit_reason])
5911                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5912         else {
5913                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5914                 vcpu->run->hw.hardware_exit_reason = exit_reason;
5915         }
5916         return 0;
5917 }
5918
5919 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5920 {
5921         if (irr == -1 || tpr < irr) {
5922                 vmcs_write32(TPR_THRESHOLD, 0);
5923                 return;
5924         }
5925
5926         vmcs_write32(TPR_THRESHOLD, irr);
5927 }
5928
5929 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
5930 {
5931         u32 exit_intr_info;
5932
5933         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
5934               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
5935                 return;
5936
5937         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5938         exit_intr_info = vmx->exit_intr_info;
5939
5940         /* Handle machine checks before interrupts are enabled */
5941         if (is_machine_check(exit_intr_info))
5942                 kvm_machine_check();
5943
5944         /* We need to handle NMIs before interrupts are enabled */
5945         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
5946             (exit_intr_info & INTR_INFO_VALID_MASK)) {
5947                 kvm_before_handle_nmi(&vmx->vcpu);
5948                 asm("int $2");
5949                 kvm_after_handle_nmi(&vmx->vcpu);
5950         }
5951 }
5952
5953 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
5954 {
5955         u32 exit_intr_info;
5956         bool unblock_nmi;
5957         u8 vector;
5958         bool idtv_info_valid;
5959
5960         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5961
5962         if (cpu_has_virtual_nmis()) {
5963                 if (vmx->nmi_known_unmasked)
5964                         return;
5965                 /*
5966                  * Can't use vmx->exit_intr_info since we're not sure what
5967                  * the exit reason is.
5968                  */
5969                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5970                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
5971                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
5972                 /*
5973                  * SDM 3: 27.7.1.2 (September 2008)
5974                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
5975                  * a guest IRET fault.
5976                  * SDM 3: 23.2.2 (September 2008)
5977                  * Bit 12 is undefined in any of the following cases:
5978                  *  If the VM exit sets the valid bit in the IDT-vectoring
5979                  *   information field.
5980                  *  If the VM exit is due to a double fault.
5981                  */
5982                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
5983                     vector != DF_VECTOR && !idtv_info_valid)
5984                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5985                                       GUEST_INTR_STATE_NMI);
5986                 else
5987                         vmx->nmi_known_unmasked =
5988                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
5989                                   & GUEST_INTR_STATE_NMI);
5990         } else if (unlikely(vmx->soft_vnmi_blocked))
5991                 vmx->vnmi_blocked_time +=
5992                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
5993 }
5994
5995 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
5996                                       u32 idt_vectoring_info,
5997                                       int instr_len_field,
5998                                       int error_code_field)
5999 {
6000         u8 vector;
6001         int type;
6002         bool idtv_info_valid;
6003
6004         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6005
6006         vmx->vcpu.arch.nmi_injected = false;
6007         kvm_clear_exception_queue(&vmx->vcpu);
6008         kvm_clear_interrupt_queue(&vmx->vcpu);
6009
6010         if (!idtv_info_valid)
6011                 return;
6012
6013         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6014
6015         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6016         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6017
6018         switch (type) {
6019         case INTR_TYPE_NMI_INTR:
6020                 vmx->vcpu.arch.nmi_injected = true;
6021                 /*
6022                  * SDM 3: 27.7.1.2 (September 2008)
6023                  * Clear bit "block by NMI" before VM entry if a NMI
6024                  * delivery faulted.
6025                  */
6026                 vmx_set_nmi_mask(&vmx->vcpu, false);
6027                 break;
6028         case INTR_TYPE_SOFT_EXCEPTION:
6029                 vmx->vcpu.arch.event_exit_inst_len =
6030                         vmcs_read32(instr_len_field);
6031                 /* fall through */
6032         case INTR_TYPE_HARD_EXCEPTION:
6033                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6034                         u32 err = vmcs_read32(error_code_field);
6035                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
6036                 } else
6037                         kvm_queue_exception(&vmx->vcpu, vector);
6038                 break;
6039         case INTR_TYPE_SOFT_INTR:
6040                 vmx->vcpu.arch.event_exit_inst_len =
6041                         vmcs_read32(instr_len_field);
6042                 /* fall through */
6043         case INTR_TYPE_EXT_INTR:
6044                 kvm_queue_interrupt(&vmx->vcpu, vector,
6045                         type == INTR_TYPE_SOFT_INTR);
6046                 break;
6047         default:
6048                 break;
6049         }
6050 }
6051
6052 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6053 {
6054         if (is_guest_mode(&vmx->vcpu))
6055                 return;
6056         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6057                                   VM_EXIT_INSTRUCTION_LEN,
6058                                   IDT_VECTORING_ERROR_CODE);
6059 }
6060
6061 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6062 {
6063         if (is_guest_mode(vcpu))
6064                 return;
6065         __vmx_complete_interrupts(to_vmx(vcpu),
6066                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6067                                   VM_ENTRY_INSTRUCTION_LEN,
6068                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
6069
6070         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6071 }
6072
6073 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6074 {
6075         int i, nr_msrs;
6076         struct perf_guest_switch_msr *msrs;
6077
6078         msrs = perf_guest_get_msrs(&nr_msrs);
6079
6080         if (!msrs)
6081                 return;
6082
6083         for (i = 0; i < nr_msrs; i++)
6084                 if (msrs[i].host == msrs[i].guest)
6085                         clear_atomic_switch_msr(vmx, msrs[i].msr);
6086                 else
6087                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6088                                         msrs[i].host);
6089 }
6090
6091 #ifdef CONFIG_X86_64
6092 #define R "r"
6093 #define Q "q"
6094 #else
6095 #define R "e"
6096 #define Q "l"
6097 #endif
6098
6099 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6100 {
6101         struct vcpu_vmx *vmx = to_vmx(vcpu);
6102
6103         if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6104                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6105                 if (vmcs12->idt_vectoring_info_field &
6106                                 VECTORING_INFO_VALID_MASK) {
6107                         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6108                                 vmcs12->idt_vectoring_info_field);
6109                         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6110                                 vmcs12->vm_exit_instruction_len);
6111                         if (vmcs12->idt_vectoring_info_field &
6112                                         VECTORING_INFO_DELIVER_CODE_MASK)
6113                                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6114                                         vmcs12->idt_vectoring_error_code);
6115                 }
6116         }
6117
6118         /* Record the guest's net vcpu time for enforced NMI injections. */
6119         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6120                 vmx->entry_time = ktime_get();
6121
6122         /* Don't enter VMX if guest state is invalid, let the exit handler
6123            start emulation until we arrive back to a valid state */
6124         if (vmx->emulation_required && emulate_invalid_guest_state)
6125                 return;
6126
6127         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6128                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6129         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6130                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6131
6132         /* When single-stepping over STI and MOV SS, we must clear the
6133          * corresponding interruptibility bits in the guest state. Otherwise
6134          * vmentry fails as it then expects bit 14 (BS) in pending debug
6135          * exceptions being set, but that's not correct for the guest debugging
6136          * case. */
6137         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6138                 vmx_set_interrupt_shadow(vcpu, 0);
6139
6140         atomic_switch_perf_msrs(vmx);
6141
6142         vmx->__launched = vmx->loaded_vmcs->launched;
6143         asm(
6144                 /* Store host registers */
6145                 "push %%"R"dx; push %%"R"bp;"
6146                 "push %%"R"cx \n\t" /* placeholder for guest rcx */
6147                 "push %%"R"cx \n\t"
6148                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
6149                 "je 1f \n\t"
6150                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
6151                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6152                 "1: \n\t"
6153                 /* Reload cr2 if changed */
6154                 "mov %c[cr2](%0), %%"R"ax \n\t"
6155                 "mov %%cr2, %%"R"dx \n\t"
6156                 "cmp %%"R"ax, %%"R"dx \n\t"
6157                 "je 2f \n\t"
6158                 "mov %%"R"ax, %%cr2 \n\t"
6159                 "2: \n\t"
6160                 /* Check if vmlaunch of vmresume is needed */
6161                 "cmpl $0, %c[launched](%0) \n\t"
6162                 /* Load guest registers.  Don't clobber flags. */
6163                 "mov %c[rax](%0), %%"R"ax \n\t"
6164                 "mov %c[rbx](%0), %%"R"bx \n\t"
6165                 "mov %c[rdx](%0), %%"R"dx \n\t"
6166                 "mov %c[rsi](%0), %%"R"si \n\t"
6167                 "mov %c[rdi](%0), %%"R"di \n\t"
6168                 "mov %c[rbp](%0), %%"R"bp \n\t"
6169 #ifdef CONFIG_X86_64
6170                 "mov %c[r8](%0),  %%r8  \n\t"
6171                 "mov %c[r9](%0),  %%r9  \n\t"
6172                 "mov %c[r10](%0), %%r10 \n\t"
6173                 "mov %c[r11](%0), %%r11 \n\t"
6174                 "mov %c[r12](%0), %%r12 \n\t"
6175                 "mov %c[r13](%0), %%r13 \n\t"
6176                 "mov %c[r14](%0), %%r14 \n\t"
6177                 "mov %c[r15](%0), %%r15 \n\t"
6178 #endif
6179                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
6180
6181                 /* Enter guest mode */
6182                 "jne .Llaunched \n\t"
6183                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
6184                 "jmp .Lkvm_vmx_return \n\t"
6185                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
6186                 ".Lkvm_vmx_return: "
6187                 /* Save guest registers, load host registers, keep flags */
6188                 "mov %0, %c[wordsize](%%"R"sp) \n\t"
6189                 "pop %0 \n\t"
6190                 "mov %%"R"ax, %c[rax](%0) \n\t"
6191                 "mov %%"R"bx, %c[rbx](%0) \n\t"
6192                 "pop"Q" %c[rcx](%0) \n\t"
6193                 "mov %%"R"dx, %c[rdx](%0) \n\t"
6194                 "mov %%"R"si, %c[rsi](%0) \n\t"
6195                 "mov %%"R"di, %c[rdi](%0) \n\t"
6196                 "mov %%"R"bp, %c[rbp](%0) \n\t"
6197 #ifdef CONFIG_X86_64
6198                 "mov %%r8,  %c[r8](%0) \n\t"
6199                 "mov %%r9,  %c[r9](%0) \n\t"
6200                 "mov %%r10, %c[r10](%0) \n\t"
6201                 "mov %%r11, %c[r11](%0) \n\t"
6202                 "mov %%r12, %c[r12](%0) \n\t"
6203                 "mov %%r13, %c[r13](%0) \n\t"
6204                 "mov %%r14, %c[r14](%0) \n\t"
6205                 "mov %%r15, %c[r15](%0) \n\t"
6206 #endif
6207                 "mov %%cr2, %%"R"ax   \n\t"
6208                 "mov %%"R"ax, %c[cr2](%0) \n\t"
6209
6210                 "pop  %%"R"bp; pop  %%"R"dx \n\t"
6211                 "setbe %c[fail](%0) \n\t"
6212               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6213                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6214                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
6215                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6216                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6217                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6218                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6219                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6220                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6221                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6222                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6223 #ifdef CONFIG_X86_64
6224                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6225                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6226                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6227                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6228                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6229                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6230                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6231                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6232 #endif
6233                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6234                 [wordsize]"i"(sizeof(ulong))
6235               : "cc", "memory"
6236                 , R"ax", R"bx", R"di", R"si"
6237 #ifdef CONFIG_X86_64
6238                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6239 #endif
6240               );
6241
6242         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6243                                   | (1 << VCPU_EXREG_RFLAGS)
6244                                   | (1 << VCPU_EXREG_CPL)
6245                                   | (1 << VCPU_EXREG_PDPTR)
6246                                   | (1 << VCPU_EXREG_SEGMENTS)
6247                                   | (1 << VCPU_EXREG_CR3));
6248         vcpu->arch.regs_dirty = 0;
6249
6250         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6251
6252         if (is_guest_mode(vcpu)) {
6253                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6254                 vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6255                 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6256                         vmcs12->idt_vectoring_error_code =
6257                                 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6258                         vmcs12->vm_exit_instruction_len =
6259                                 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6260                 }
6261         }
6262
6263         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
6264         vmx->loaded_vmcs->launched = 1;
6265
6266         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6267         trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6268
6269         vmx_complete_atomic_exit(vmx);
6270         vmx_recover_nmi_blocking(vmx);
6271         vmx_complete_interrupts(vmx);
6272 }
6273
6274 #undef R
6275 #undef Q
6276
6277 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6278 {
6279         struct vcpu_vmx *vmx = to_vmx(vcpu);
6280
6281         free_vpid(vmx);
6282         free_nested(vmx);
6283         free_loaded_vmcs(vmx->loaded_vmcs);
6284         kfree(vmx->guest_msrs);
6285         kvm_vcpu_uninit(vcpu);
6286         kmem_cache_free(kvm_vcpu_cache, vmx);
6287 }
6288
6289 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6290 {
6291         int err;
6292         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6293         int cpu;
6294
6295         if (!vmx)
6296                 return ERR_PTR(-ENOMEM);
6297
6298         allocate_vpid(vmx);
6299
6300         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6301         if (err)
6302                 goto free_vcpu;
6303
6304         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6305         err = -ENOMEM;
6306         if (!vmx->guest_msrs) {
6307                 goto uninit_vcpu;
6308         }
6309
6310         vmx->loaded_vmcs = &vmx->vmcs01;
6311         vmx->loaded_vmcs->vmcs = alloc_vmcs();
6312         if (!vmx->loaded_vmcs->vmcs)
6313         &